diff --git a/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/README.md b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/README.md new file mode 100644 index 0000000000..ac177659af --- /dev/null +++ b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/README.md @@ -0,0 +1,95 @@ +# Record: Scylla Tokenizer + Full GPTQ + XSA-all + FA3 + +**val_bpb: 0.9485** (3-seed mean, std 0.0008) | **~15.6 MB** | 8×H100 SXM | No TTT + +## Results (8×H100 SXM) + +| Seed | Steps | ms/step | Sliding BPB (s64) | +|------|-------|---------|--------------------| +| 1337 | 6,716 | 87.9 | 0.9491 | +| 42 | — | ~88 | **0.9476** | +| 2025 | — | ~88 | 0.9489 | +| **Mean ± Std** | | | **0.9485 ± 0.0008** | + +## vs Competition + +| Submission | BPB | Improvement | +|-----------|-----|-------------| +| PR #1143 (Scylla, old stack) | 1.0806 | **-0.1321 (-12.2%)** | +| PR #1089 (Turbo-Muon) | 1.1086 | -0.1601 | +| Merged SOTA (PR #549) | 1.1194 | -0.1709 | + +## What's New + +This submission combines the Scylla tokenizer (PR #1143) with the modern training stack (PR #1060), achieving a result far better than either alone. + +### Tokenizer: Scylla (998 tokens) +- TokenMonster-derived vocabulary, pruned from `english-1024-clean-v1` +- Created by @simon-marcus through iterative autoresearch (PR #1143) +- 998 active tokens (vs 1024 for SentencePiece) +- Better byte-per-token efficiency via ungreedy multi-branch tokenization +- Retokenized FineWeb: 194 train shards (~19.4B tokens) + 1 val shard + +### Training Stack (PR #1060 base) +- **Full Hessian GPTQ** — Cholesky error compensation, 64-batch calibration in 6.7s +- **XSA on all 11 layers** — exclusive self-attention everywhere +- **Coprime-stride multi-shard loader** — diverse batches across 194 shards +- **FlashAttention 3** — Hopper native kernels (pre-built wheel) +- **Parallel Muon** + Parameter Banking — 3-phase overlapped optimizer + +### Why It's Better Than #1143 Alone +PR #1143 used the old SOTA stack (PR #549 base) which lacks: +- Full GPTQ (used GPTQ-lite → worse quantization) +- XSA on all layers (used last 4 → less cross-position mixing) +- Coprime data loader (sequential loading → less batch diversity) +- More training data (we used 194 shards vs their 79) + +### No TTT Needed +TTT was tested and found neutral (0.9491 with TTT, 0.9491 without). Full GPTQ eliminates the need for test-time adaptation. + +## Architecture + +- 11L, 512d, 8H/4KV (GQA), MLP 3× LeakyReLU(0.5)² +- XSA on all 11 layers, BigramHash(2816×112), SmearGate +- Partial RoPE (16d), LN Scale 1/√(l+1) +- Shared ValueEmbedding (dim=128, layers 9-10) +- EMA (decay=0.997) + Tight SWA (every 50 steps) +- Full Hessian GPTQ int6 + LZMA compression + +## Timing + +| Phase | Time | +|-------|------| +| Training (6,716 steps @ 88ms) | 591s | +| GPTQ calibration | 6.7s | +| Sliding window eval (stride=64) | 92s | + +## Reproduction + +```bash +# Install +pip install torch==2.9.1 --index-url https://download.pytorch.org/whl/cu128 +pip install "https://download.pytorch.org/whl/cu128/flash_attn_3-3.0.0-cp39-abi3-manylinux_2_28_x86_64.whl" +pip install sentencepiece huggingface-hub datasets numpy tqdm tokenmonster + +# Retokenize FineWeb with Scylla vocab (once, ~90 min) +python3 retokenize.py --vocab candidate.vocab --output-dir data/datasets/fineweb10B_scylla + +# Train +SEED=1337 DATA_PATH=./data/datasets/fineweb10B_scylla \ +TOKENIZER_PATH=./candidate.vocab TOKENIZER_META_PATH=./candidate.meta.npz \ +VOCAB_SIZE=998 XSA_LAST_N=11 USE_GPTQ=1 GPTQ_RESERVE_MS=9000 TTT_ENABLED=0 \ +BIGRAM_VOCAB_SIZE=2816 BIGRAM_DIM=112 \ +torchrun --standalone --nproc_per_node=8 train_gpt.py +``` + +## Included Files +- `train_gpt.py` — Training script (PR #1060 + tokenizer metadata loading) +- `candidate.vocab` — Scylla tokenizer (998 tokens) +- `candidate.meta.npz` — Per-token byte accounting metadata +- `train_seed{1337,42,2025}.log` — Training logs for all 3 seeds + +## Credits +- **Scylla tokenizer**: @simon-marcus (PR #1143) +- **Training stack**: @resouer (PR #1060), @abaybektursun (PR #549) +- **Retokenization pipeline**: Built for this submission diff --git a/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/candidate.meta.npz b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/candidate.meta.npz new file mode 100644 index 0000000000..580a1e39e5 Binary files /dev/null and b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/candidate.meta.npz differ diff --git a/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/candidate.vocab b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/candidate.vocab new file mode 100644 index 0000000000..a2b7009ad3 Binary files /dev/null and b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/candidate.vocab differ diff --git a/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/submission.json b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/submission.json new file mode 100644 index 0000000000..1f74c0ffe0 --- /dev/null +++ b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/submission.json @@ -0,0 +1,9 @@ +{ + "name": "Scylla tokenizer + Full GPTQ + XSA-all + FA3", + "val_bpb": 0.9485, + "bytes_total": 15600000, + "blurb": "Scylla TokenMonster tokenizer (998 tokens) + PR #1060 stack (Full Hessian GPTQ, XSA on all 11 layers, coprime-stride loader) + FA3 Hopper attention. 3-seed mean 0.9485 bpb (std 0.0008). No TTT needed. 87.9ms/step, 6716 steps.", + "author": "icryo", + "github_id": "icryo", + "date": "2026-03-31" +} diff --git a/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_gpt.py b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_gpt.py new file mode 100644 index 0000000000..6e8f2f90d0 --- /dev/null +++ b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_gpt.py @@ -0,0 +1,2173 @@ +from __future__ import annotations +import copy +import glob +import io +import lzma +import math +import os +import random +import subprocess +import sys +import time +import uuid +from pathlib import Path +import numpy as np +import sentencepiece as spm +import torch +import torch.distributed as dist +import torch.nn.functional as F +from torch import Tensor, nn +from torch.nn.parallel import DistributedDataParallel as DDP +from flash_attn_interface import flash_attn_func as flash_attn_3_func +class Hyperparameters: + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp1024") + train_files = os.path.join(data_path, "fineweb_train_*.bin") + val_files = os.path.join(data_path, "fineweb_val_*.bin") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_1024_bpe.model") + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + seed = int(os.environ.get("SEED", 1337)) + val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 3500)) + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786_432)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 1.5)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 1024)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 3.0)) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.035)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.025)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.025)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.99)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + swa_enabled = bool(int(os.environ.get("SWA_ENABLED", "1"))) + swa_every = int(os.environ.get("SWA_EVERY", 50)) + muon_wd = float(os.environ.get("MUON_WD", 0.04)) + adam_wd = float(os.environ.get("ADAM_WD", 0.04)) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 2048)) + bigram_dim = int(os.environ.get("BIGRAM_DIM", 128)) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 4)) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + ve_enabled = bool(int(os.environ.get("VE_ENABLED", "1"))) + ve_dim = int(os.environ.get("VE_DIM", 128)) + ve_layers = os.environ.get("VE_LAYERS", "9,10") + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "0"))) + ttt_lr = float(os.environ.get("TTT_LR", 0.002)) + ttt_epochs = int(os.environ.get("TTT_EPOCHS", 3)) + ttt_chunk_tokens = int(os.environ.get("TTT_CHUNK_TOKENS", 32768)) + ttt_freeze_blocks = int(os.environ.get("TTT_FREEZE_BLOCKS", 2)) + ttt_momentum = float(os.environ.get("TTT_MOMENTUM", 0.9)) + ttt_batch_seqs = int(os.environ.get("TTT_BATCH_SEQS", 32)) + ttt_grad_clip = float(os.environ.get("TTT_GRAD_CLIP", 1.0)) + negative_slope = float(os.environ.get("NEGATIVE_SLOPE", 0.5)) + use_gptq = bool(int(os.environ.get("USE_GPTQ", "0"))) + gptq_calib_samples = int(os.environ.get("GPTQ_CALIB_SAMPLES", "64")) + gptq_reserve_ms = float(os.environ.get("GPTQ_RESERVE_MS", "14000")) + quant_clip_range = int(os.environ.get("QUANT_CLIP_RANGE", 31)) + tokenizer_meta_path = os.environ.get("TOKENIZER_META_PATH", "") + tokenizer_meta_validate = bool(int(os.environ.get("TOKENIZER_META_VALIDATE", "0"))) + +# --- Batched Newton-Schulz orthogonalization --- + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 5, eps: float = 1e-7) -> Tensor: + """Batched Newton-Schulz orthogonalization. G: (B,M,N) or (M,N).""" + a, b, c = (3.4445, -4.7750, 2.0315) + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + for _ in range(steps): + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + +# --- Parallel Muon optimizer --- + +class Muon(torch.optim.Optimizer): + """Parallel Muon: post-backward reduce-scatter -> local NS5 -> all-gather. + + No DDP for bank params. After backward, this optimizer: + 1. Launches async reduce-scatter for all banks (biggest first) + 2. Returns control so Adam can step on small params while RS is in-flight + 3. Waits for each RS, runs local NS5 on the shard, launches async all-gather + 4. Each all-gather overlaps with next bank's NS5 + """ + def __init__(self, params, lr: float, momentum: float, backend_steps: int, + nesterov: bool = True, weight_decay: float = 0.0): + super().__init__( + params, + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, + nesterov=nesterov, weight_decay=weight_decay), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + 'p': p, + 'B': B, + 'padded_grad': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard_mom': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'full_update': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'scale': max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + # Sort by size descending -- launch biggest reduce-scatters first + self._bank_meta.sort(key=lambda m: -m['p'].numel()) + self._built = True + + def launch_reduce_scatters(self): + """Phase 1: launch async reduce-scatter for all banks. Call right after backward.""" + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m['p'] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m['padded_grad'] + pg[:m['B']].copy_(p.grad.bfloat16()) + if pg.shape[0] > m['B']: + pg[m['B']:].zero_() + fut = dist.reduce_scatter_tensor(m['shard'], pg, op=dist.ReduceOp.AVG, async_op=True) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + """Phase 3: wait for RS, local NS5, all-gather. Call AFTER Adam steps.""" + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + + if not self._built: + self._build() + + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + + prev_ag_handle = None + prev_m = None + + sharded = self._distributed and hasattr(self, '_rs_futures') + + for i, m in enumerate(self._bank_meta): + p = m['p'] + if p.grad is None: + continue + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if sharded and self._rs_futures[i] is not None: + self._rs_futures[i].wait() + g = m['shard'] + buf = m['shard_mom'] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m['full_update'], update, async_op=True) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m['scale']) + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if hasattr(self, '_rs_futures'): + del self._rs_futures + + return loss + +# --- Tokenizer evaluation helpers --- + +def build_sentencepiece_luts( + sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device +) -> tuple[Tensor, Tensor, Tensor]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) + +TOKENIZER_META_FORMAT_VERSION = 1 +TOKENIZER_META_SUFFIX = ".meta.npz" + + +def _derive_tokenizer_meta_path(tokenizer_path: str) -> Path: + tokenizer = Path(tokenizer_path) + if tokenizer.suffix == ".model": + return tokenizer.with_suffix(TOKENIZER_META_SUFFIX) + return tokenizer.with_name(tokenizer.name + TOKENIZER_META_SUFFIX) + + +def build_sentencepiece_luts_np( + sp: spm.SentencePieceProcessor, vocab_size: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return base_bytes_np, has_leading_space_np, is_boundary_token_np + + +def load_tokenizer_meta_luts_np( + meta_path: Path, vocab_size: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray, dict[str, object]]: + def _scalar(value): + arr = np.asarray(value) + if arr.ndim == 0: + return arr.item() + first = arr.reshape(-1)[0] + return first.item() if hasattr(first, "item") else first + + with np.load(meta_path, allow_pickle=False) as data: + format_version = int(_scalar(data["format_version"])) + if format_version != TOKENIZER_META_FORMAT_VERSION: + raise ValueError( + f"Unsupported tokenizer meta format_version={format_version} " + f"expected={TOKENIZER_META_FORMAT_VERSION}" + ) + meta_vocab_size = int(_scalar(data["vocab_size"])) + tokenizer_kind = str(_scalar(data["tokenizer_kind"])) + source_model_name = str(_scalar(data["source_model_name"])) + base_bytes_np = np.asarray(data["base_bytes"], dtype=np.int16) + has_leading_space_np = np.asarray(data["has_leading_space"], dtype=np.bool_) + is_boundary_token_np = np.asarray(data["is_boundary_token"], dtype=np.bool_) + table_size = max(meta_vocab_size, vocab_size) + if base_bytes_np.shape[0] < table_size: + padded_base_bytes = np.zeros((table_size,), dtype=np.int16) + padded_has_leading_space = np.zeros((table_size,), dtype=np.bool_) + padded_is_boundary = np.ones((table_size,), dtype=np.bool_) + padded_base_bytes[: base_bytes_np.shape[0]] = base_bytes_np + padded_has_leading_space[: has_leading_space_np.shape[0]] = has_leading_space_np + padded_is_boundary[: is_boundary_token_np.shape[0]] = is_boundary_token_np + base_bytes_np = padded_base_bytes + has_leading_space_np = padded_has_leading_space + is_boundary_token_np = padded_is_boundary + metadata = { + "format_version": format_version, + "tokenizer_kind": tokenizer_kind, + "source_model_name": source_model_name, + "vocab_size": meta_vocab_size, + "meta_path": str(meta_path), + } + return base_bytes_np, has_leading_space_np, is_boundary_token_np, metadata + + +def load_tokenizer_luts( + tokenizer_path: str, + tokenizer_meta_path: str, + vocab_size: int, + device: torch.device, + *, + validate_meta: bool = False, +) -> tuple[tuple[Tensor, Tensor, Tensor], dict[str, object]]: + meta_path = ( + Path(tokenizer_meta_path) if tokenizer_meta_path + else _derive_tokenizer_meta_path(tokenizer_path) + ) + if meta_path.exists(): + base_bytes_np, has_leading_space_np, is_boundary_token_np, metadata = ( + load_tokenizer_meta_luts_np(meta_path, vocab_size) + ) + if validate_meta and str(tokenizer_path).endswith(".model"): + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + sp_luts = build_sentencepiece_luts_np(sp, vocab_size) + if not ( + np.array_equal(base_bytes_np, sp_luts[0]) + and np.array_equal(has_leading_space_np, sp_luts[1]) + and np.array_equal(is_boundary_token_np, sp_luts[2]) + ): + raise ValueError(f"Tokenizer metadata mismatch for {meta_path}") + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ), metadata + if not str(tokenizer_path).endswith(".model"): + raise FileNotFoundError(f"TOKENIZER_META_PATH does not exist: {meta_path}") + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + return build_sentencepiece_luts(sp, vocab_size, device), { + "tokenizer_kind": "sentencepiece", + "source_model_name": str(tokenizer_path), + "vocab_size": int(sp.vocab_size()), + "meta_path": None, + "fallback": True, + } + +def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = ((tokens.numel() - 1) // seq_len) * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] +def eval_val( + args: Hyperparameters, + model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + grad_accum_steps: int, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + seq_len = eval_seq_len or args.train_seq_len + local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + "VAL_BATCH_SIZE must provide at least one sequence per rank; " + f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " + f"GRAD_ACCUM_STEPS={grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + seq_start = (total_seqs * rank) // world_size + seq_end = (total_seqs * (rank + 1)) // world_size + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + model.eval() + with torch.inference_mode(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + batch_loss = model(x, y).detach() + batch_token_count = float(y.numel()) + val_loss_sum += batch_loss.to(torch.float64) * batch_token_count + val_token_count += batch_token_count + prev_ids = x.reshape(-1) + tgt_ids = y.reshape(-1) + token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + val_loss = val_loss_sum / val_token_count + bits_per_token = val_loss.item() / math.log(2.0) + tokens_per_byte = val_token_count.item() / val_byte_count.item() + model.train() + return float(val_loss.item()), float(bits_per_token * tokens_per_byte) + +# --- Quantization helpers --- + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,smear,dtg_gate,ve_layer_scales,ve_shared.scale,attn_gate,vr_lambda", + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_FP32_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "INT8_KEEP_FLOAT_FP32_NAME_PATTERNS", + ",".join(CONTROL_TENSOR_NAME_PATTERNS), + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_MAX_NUMEL = 65_536 +INT8_KEEP_FLOAT_STORE_DTYPE = torch.float16 +INT8_PER_ROW_SCALE_DTYPE = torch.float16 +INT8_CLIP_PERCENTILE = 99.99984 +INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 +def tensor_nbytes(t: Tensor) -> int: + return int(t.numel()) * int(t.element_size()) +def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: + if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): + return t.float().contiguous() + if t.dtype in {torch.float32, torch.bfloat16}: + passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") + return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() + return t +def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + clip_abs = ( + torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) + if t32.numel() + else torch.empty((t32.shape[0],), dtype=torch.float32) + ) + clipped = torch.maximum(torch.minimum(t32, clip_abs[:, None]), -clip_abs[:, None]) + scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) + q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() + return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() + clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 + scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) + q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() + return q, scale +def quantize_state_dict_int8(state_dict: dict[str, Tensor]): + quantized: dict[str, Tensor] = {} + scales: dict[str, Tensor] = {} + dtypes: dict[str, str] = {} + passthrough: dict[str, Tensor] = {} + passthrough_orig_dtypes: dict[str, str] = {} + qmeta: dict[str, dict[str, object]] = {} + stats = dict.fromkeys( + ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), + 0, + ) + for name, tensor in state_dict.items(): + t = tensor.detach().to("cpu").contiguous() + stats["param_count"] += int(t.numel()) + stats["num_tensors"] += 1 + stats["baseline_tensor_bytes"] += tensor_nbytes(t) + if not t.is_floating_point(): + stats["num_nonfloat_tensors"] += 1 + passthrough[name] = t + stats["int8_payload_bytes"] += tensor_nbytes(t) + continue + if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: + kept = keep_float_tensor(name, t, passthrough_orig_dtypes) + passthrough[name] = kept + stats["int8_payload_bytes"] += tensor_nbytes(kept) + continue + stats["num_float_tensors"] += 1 + q, s = quantize_float_tensor(t) + if s.ndim > 0: + qmeta[name] = {"scheme": "per_row", "axis": 0} + quantized[name] = q + scales[name] = s + dtypes[name] = str(t.dtype).removeprefix("torch.") + stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) + obj: dict[str, object] = { + "__quant_format__": "int8_clean_per_row_v1", + "quantized": quantized, + "scales": scales, + "dtypes": dtypes, + "passthrough": passthrough, + } + if qmeta: + obj["qmeta"] = qmeta + if passthrough_orig_dtypes: + obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes + return obj, stats +def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + qmeta = obj.get("qmeta", {}) + passthrough_orig_dtypes = obj.get("passthrough_orig_dtypes", {}) + for name, q in obj["quantized"].items(): + dtype = getattr(torch, obj["dtypes"][name]) + s = obj["scales"][name] + if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: + s = s.to(dtype=torch.float32) + out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() + else: + scale = float(s.item()) + out[name] = (q.float() * scale).to(dtype=dtype).contiguous() + for name, t in obj["passthrough"].items(): + out_t = t.detach().to("cpu").contiguous() + orig_dtype = passthrough_orig_dtypes.get(name) + if isinstance(orig_dtype, str): + out_t = out_t.to(dtype=getattr(torch, orig_dtype)).contiguous() + out[name] = out_t + return out + +# --- Data loading --- + +def load_data_shard(file: Path) -> Tensor: + header_bytes = 256 * np.dtype(" int: + key = str(file) + cached = _SHARD_NTOKENS_CACHE.get(key) + if cached is not None: + return cached + header = np.fromfile(file, dtype=" np.memmap: + key = str(file) + mm = _MMAP_CACHE.get(key) + if mm is not None: + return mm + n = _read_num_tokens(file) + mm = np.memmap(file, mode="r", dtype=" int: + if n <= 1: + return 1 + while True: + s = int(self._rng.integers(1, n)) + if math.gcd(s, n) == 1: + return s + def _reset_cursor(self, si: int, seq_len: int) -> None: + nt = int(self._num_tokens[si]) + max_phase = min(seq_len - 1, max(0, nt - seq_len - 1)) + phase = int(self._rng.integers(max_phase + 1)) if max_phase > 0 else 0 + bc = (nt - 1 - phase) // seq_len + self._cursor_phase[si] = phase + self._cursor_block_count[si] = bc + self._cursor_next[si] = 0 + self._cursor_start[si] = int(self._rng.integers(bc)) if bc > 1 else 0 + self._cursor_stride[si] = self._pick_coprime_stride(bc) + self._cursor_init[si] = True + def _ensure_cursor(self, si: int, seq_len: int) -> None: + if not self._cursor_init[si] or self._cursor_next[si] >= self._cursor_block_count[si]: + self._reset_cursor(si, seq_len) + def _take_from_shard(self, si: int, seq_len: int, count: int, out: list[tuple[int, int]]) -> None: + rem = count + while rem > 0: + self._ensure_cursor(si, seq_len) + bc = int(self._cursor_block_count[si]) + ni = int(self._cursor_next[si]) + take = min(rem, bc - ni) + phase = int(self._cursor_phase[si]) + start = int(self._cursor_start[si]) + stride = int(self._cursor_stride[si]) + for j in range(take): + bi = (start + (ni + j) * stride) % bc + out.append((si, phase + bi * seq_len)) + self._cursor_next[si] = ni + take + rem -= take + def _init_pipeline(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> None: + local_tokens = global_tokens // (self.world_size * grad_accum_steps) + num_seqs = local_tokens // seq_len + global_num_seqs = num_seqs * self.world_size + self._cfg = (local_tokens, seq_len, num_seqs, global_num_seqs) + bbc = (self._num_tokens - 1) // seq_len + eligible = bbc > 0 + self._eligible_shards = np.nonzero(eligible)[0].astype(np.int64) + self._base_block_counts = bbc[self._eligible_shards].astype(np.int64) + def _sample_global_windows(self) -> list[tuple[int, int]]: + assert self._cfg is not None and self._eligible_shards is not None + _, seq_len, _, gns = self._cfg + ec = int(self._eligible_shards.size) + progress = min(self._batches_built / 1800.0, 1.0) + remaining = np.empty(ec, dtype=np.float64) + for i, si in enumerate(self._eligible_shards.tolist()): + if self._cursor_init[si]: + r = int(self._cursor_block_count[si]) - int(self._cursor_next[si]) + remaining[i] = float(max(r, 1)) + else: + remaining[i] = float(self._base_block_counts[i]) + alpha = 0.90 - 0.40 * progress + weights = np.power(remaining, alpha) + ws = float(weights.sum()) + if not np.isfinite(ws) or ws <= 0.0: + weights = np.ones(ec, dtype=np.float64) + ws = float(weights.sum()) + probs = weights / ws + low = min(max(8, self.world_size), ec, gns) + high = min(max(32, self.world_size * 8), ec, gns) + mix = max(1, min(int(round(low + progress * (high - low))), ec, gns)) + cp = self._rng.choice(ec, size=mix, replace=False, p=probs) + cs = self._eligible_shards[cp] + cpr = probs[cp].copy() + cpr /= cpr.sum() + counts = np.ones(mix, dtype=np.int64) + extra = gns - mix + if extra > 0: + counts += self._rng.multinomial(extra, cpr).astype(np.int64) + perm = self._rng.permutation(mix) + cs, counts = cs[perm], counts[perm] + buckets: list[list[tuple[int, int]]] = [] + for si, cnt in zip(cs.tolist(), counts.tolist()): + b: list[tuple[int, int]] = [] + self._take_from_shard(int(si), seq_len, int(cnt), b) + if b: + if len(b) > 1: + bp = self._rng.permutation(len(b)) + b = [b[int(k)] for k in bp.tolist()] + buckets.append(b) + windows: list[tuple[int, int]] = [] + active = [i for i, bk in enumerate(buckets) if bk] + while active: + order = self._rng.permutation(len(active)) + new_active: list[int] = [] + for oi in order.tolist(): + bi = active[oi] + if buckets[bi]: + windows.append(buckets[bi].pop()) + if buckets[bi]: + new_active.append(bi) + active = new_active + return windows + def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: + if self._cfg is None: + self._init_pipeline(global_tokens, seq_len, grad_accum_steps) + _, _, num_seqs, gns = self._cfg + gw = self._sample_global_windows() + local_w = gw[self.rank::self.world_size] + x = torch.empty((num_seqs, seq_len), dtype=torch.int64) + y = torch.empty((num_seqs, seq_len), dtype=torch.int64) + for slot, (si, pos) in enumerate(local_w): + mm = _get_shard_memmap(self.files[si]) + window = torch.as_tensor(np.array(mm[pos:pos + seq_len + 1], dtype=np.int64)) + x[slot] = window[:-1] + y[slot] = window[1:] + self._batches_built += 1 + return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) + +# --- Transformer modules --- + +class RMSNorm(nn.Module): + def __init__(self, eps: float | None = None): + super().__init__() + self.eps = eps + def forward(self, x: Tensor) -> Tensor: + return F.rms_norm(x, (x.size(-1),), eps=self.eps) +class CastedLinear(nn.Linear): + def forward(self, x: Tensor) -> Tensor: + w = self.weight.to(x.dtype) + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) +def restore_low_dim_params_to_fp32(module: nn.Module) -> None: + with torch.no_grad(): + for name, param in module.named_parameters(): + if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: + param.data = param.data.float() +class Rotary(nn.Module): + def __init__(self, dim: int, base: float = 10000.0, train_seq_len: int = 1024, rope_dims: int = 0): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / (base ** (torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached: Tensor | None = None + self._sin_cached: Tensor | None = None + def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached != seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * (scale ** (rd / (rd - 2))) + inv_freq = 1.0 / (new_base ** (torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd)) + else: + inv_freq = self.inv_freq.to(device) + t = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor, rope_dims: int = 0) -> Tensor: + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + +class CausalSelfAttention(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + rope_base: float, + qk_gain_init: float, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + # No CastedLinear -- weights come from banks + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 # set by GPT.__init__ for partial RoPE + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False # set by GPT.__init__ for deep layers only + def _xsa_efficient(self, y: Tensor, v: Tensor) -> Tensor: + """Efficient XSA: subtract self-value projection via GQA-aware reshape (no repeat_interleave). + y: [B, T, H, D], v: [B, T, Hkv, D]. H must be divisible by Hkv.""" + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) # [B, T, Hkv, group, D] + vn = F.normalize(v, dim=-1).unsqueeze(-2) # [B, T, Hkv, 1, D] -- broadcast ready + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, v_embed: Tensor | None = None) -> Tensor: + if getattr(self, '_save_gptq', False): + self._gptq_qkv_in = x.detach() + bsz, seqlen, dim = x.shape + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, v_w.to(x.dtype)) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + y = flash_attn_3_func(q, k, v, causal=True) + if self.use_xsa: + y = self._xsa_efficient(y, v) + y = y.reshape(bsz, seqlen, dim) + if getattr(self, '_save_gptq', False): + self._gptq_o_in = y.detach() + return F.linear(y, out_w.to(x.dtype)) + +class SmearGate(nn.Module): + def __init__(self, dim: int): + super().__init__() + self.gate = nn.Parameter(torch.zeros(dim, dtype=torch.float32)) + def forward(self, x: Tensor) -> Tensor: + g = torch.sigmoid(self.gate.to(dtype=x.dtype))[None, None, :] + x_prev = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) + return (1 - g) * x + g * x_prev + +class BigramHashEmbedding(nn.Module): + def __init__(self, bigram_vocab_size: int, bigram_dim: int, model_dim: int): + super().__init__() + self.bigram_vocab_size = bigram_vocab_size + self.embed = nn.Embedding(bigram_vocab_size, bigram_dim) + nn.init.zeros_(self.embed.weight) + self.proj = CastedLinear(bigram_dim, model_dim, bias=False) if bigram_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.05, dtype=torch.float32)) + def bigram_hash(self, tokens: Tensor) -> Tensor: + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., 0] = mod + out[..., 1:] = torch.bitwise_xor(36313 * t[..., 1:], 27191 * t[..., :-1]) % mod + return out.long() + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(self.bigram_hash(token_ids)) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class ValueEmbedding(nn.Module): + """Reinject token identity into attention values at specific layers. + Each table maps vocab tokens to a low-dim embedding, projected to model_dim.""" + def __init__(self, vocab_size: int, ve_dim: int, model_dim: int): + super().__init__() + self.embed = nn.Embedding(vocab_size, ve_dim) + nn.init.normal_(self.embed.weight, std=0.01) + self.proj = CastedLinear(ve_dim, model_dim, bias=False) if ve_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.1, dtype=torch.float32)) + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(token_ids) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class MLP(nn.Module): + def __init__(self, dim: int, mlp_mult: int, neg_slope: float = 0.5): + super().__init__() + self.neg_slope = neg_slope + # No CastedLinear -- weights come from banks + def forward(self, x: Tensor, up_w: Tensor, down_w: Tensor) -> Tensor: + if getattr(self, '_save_gptq', False): + self._gptq_up_in = x.detach() + x = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=self.neg_slope) + x2 = x.square() + if getattr(self, '_save_gptq', False): + self._gptq_down_in = x2.detach() + return F.linear(x2, down_w.to(x.dtype)) + +class Block(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + rope_base: float, + qk_gain_init: float, + layer_idx: int = 0, + ln_scale: bool = False, + neg_slope: float = 0.5, + ): + super().__init__() + self.layer_idx = layer_idx + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = MLP(dim, mlp_mult, neg_slope=neg_slope) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + def forward(self, x: Tensor, x0: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, up_w: Tensor, down_w: Tensor, v_embed: Tensor | None = None) -> Tensor: + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, q_w, k_w, v_w, out_w, v_embed=v_embed) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + mlp_out = self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + return x_out + mlp_out + +class GPT(nn.Module): + def __init__( + self, + vocab_size: int, + num_layers: int, + model_dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + tie_embeddings: bool, + tied_embed_init_std: float, + logit_softcap: float, + rope_base: float, + qk_gain_init: float, + bigram_vocab_size: int = 0, + bigram_dim: int = 128, + xsa_last_n: int = 0, + rope_dims: int = 0, + ln_scale: bool = False, + ve_enabled: bool = False, + ve_dim: int = 128, + ve_layers: str = "9,10", + neg_slope: float = 0.5, + ): + super().__init__() + self._ve_target_dim = num_kv_heads * (model_dim // num_heads) # kv_dim for value projection + if logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") + self.tie_embeddings = tie_embeddings + self.tied_embed_init_std = tied_embed_init_std + self.logit_softcap = logit_softcap + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + # Parameter banks: contiguous 3D tensors for batched optimizer + head_dim = model_dim // num_heads + kv_dim = num_kv_heads * head_dim + mlp_dim = int(mlp_mult * model_dim) + self.num_layers = num_layers + self.qo_bank = nn.Parameter(torch.empty(2 * num_layers, model_dim, model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * num_layers, kv_dim, model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(num_layers, mlp_dim, model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(num_layers, model_dim, mlp_dim)) + self.blocks = nn.ModuleList( + [ + Block( + model_dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + layer_idx=i, + ln_scale=ln_scale, + neg_slope=neg_slope, + ) + for i in range(num_layers) + ] + ) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + kv_dim_ve = self._ve_target_dim + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim_ve) + self.ve_layer_scales = nn.ParameterList( + [nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices] + ) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.value_embeds = nn.ModuleList() # keep empty for compat + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + if self.lm_head is not None: + self.lm_head._zero_init = True + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + self._init_weights() + def _init_weights(self) -> None: + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + # Init banks: orthogonal, with proj layers scaled down and out/down zero-init + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) # Q + nn.init.zeros_(self.qo_bank.data[n + i]) # Out (zero init) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) # K + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) # V + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) # MLP up + nn.init.zeros_(self.mlp_down_bank.data[i]) # MLP down (zero init) + # Scale proj layers (out_proj and mlp_down are "proj" layers) + self.qo_bank.data[n + i].mul_(proj_scale) + self.mlp_down_bank.data[i].mul_(proj_scale) + # Init remaining nn.Linear modules (bigram proj, lm_head) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif module.weight.ndim == 2 and module.weight.shape[0] >= 64 and module.weight.shape[1] >= 64: + nn.init.orthogonal_(module.weight, gain=1.0) + def _get_ve(self, layer_idx: int, input_ids: Tensor, ve_cache: dict | None = None) -> Tensor | None: + """Get value embedding for a specific layer using shared table + per-layer scale.""" + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if ve_cache is not None and 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_base = ve_cache['ve'] if ve_cache is not None else self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_base * self.ve_layer_scales[ve_idx].to(dtype=ve_base.dtype) + def forward(self, input_ids: Tensor, target_ids: Tensor) -> Tensor: + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + if self.tie_embeddings: + logits_proj = F.linear(x_flat, self.tok_emb.weight) + else: + if self.lm_head is None: + raise RuntimeError("lm_head is required when tie_embeddings=False") + logits_proj = self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy(logits.float(), targets, reduction="mean") + def forward_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve) + x = self.final_norm(x) + if self.tie_embeddings: + logits_proj = F.linear(x, self.tok_emb.weight) + else: + logits_proj = self.lm_head(x) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + +# --- Sliding window evaluation --- + +def eval_val_sliding( + args: Hyperparameters, + base_model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + stride: int, + batch_seqs: int = 32, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + """Sliding window evaluation: each token scored with maximum context.""" + seq_len = eval_seq_len or args.train_seq_len + total_tokens = val_tokens.numel() - 1 + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= 1] + total_windows = len(window_starts) + my_s = (total_windows * rank) // world_size + my_e = (total_windows * (rank + 1)) // world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + base_model.eval() + compiled_logits = torch.compile(base_model.forward_logits, dynamic=False, fullgraph=True) + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk[:-1] + y_batch[i, :wlen] = chunk[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = compiled_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), + reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt = y_batch[i, s:wlen] + prev = x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + val_loss = (loss_sum / token_count).item() + bits_per_token = val_loss / math.log(2.0) + tokens_per_byte = token_count.item() / byte_count.item() + base_model.train() + return val_loss, bits_per_token * tokens_per_byte + + +def eval_val_sliding_ttt( + args: Hyperparameters, base_model: nn.Module, rank: int, world_size: int, + device: torch.device, val_tokens: Tensor, base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, is_boundary_token_lut: Tensor, + stride: int, batch_seqs: int = 32, log0=print, +) -> tuple[float, float]: + """Legal score-first TTT (PR #461 recipe): score each chunk with sliding windows, + then train on it. Every token scored BEFORE any update that could use it.""" + seq_len = args.train_seq_len + total_tokens = val_tokens.numel() - 1 + ttt_chunk = args.ttt_chunk_tokens + + # Pre-compute all window starts + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= stride or ws == 0] + + # Assign each window to a chunk based on the first token it scores + num_chunks = (total_tokens + ttt_chunk - 1) // ttt_chunk + chunk_windows: list[list[int]] = [[] for _ in range(num_chunks)] + for ws in window_starts: + end = min(ws + seq_len, total_tokens) + wlen = end - ws + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_start = ws + s + ci = min(scored_start // ttt_chunk, num_chunks - 1) + chunk_windows[ci].append(ws) + + log0(f"ttt_sliding:start chunks={num_chunks} chunk_tokens={ttt_chunk} " + f"total_windows={len(window_starts)} stride={stride} " + f"ttt_lr={args.ttt_lr} ttt_epochs={args.ttt_epochs} " + f"freeze_blocks={args.ttt_freeze_blocks}") + + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + + # Freeze first N blocks + frozen_block_ids = set(range(min(args.ttt_freeze_blocks, len(base_model.blocks)))) + ttt_params = [] + for name, p in base_model.named_parameters(): + freeze = False + for bi in frozen_block_ids: + if f"blocks.{bi}." in name: + freeze = True + break + if freeze: + p.requires_grad_(False) + else: + p.requires_grad_(True) + ttt_params.append(p) + + log0(f"ttt_sliding:params unfrozen={sum(p.numel() for p in ttt_params)} " + f"frozen={sum(p.numel() for p in base_model.parameters() if not p.requires_grad)}") + + optimizer = torch.optim.SGD(ttt_params, lr=args.ttt_lr, momentum=args.ttt_momentum) + t0 = time.perf_counter() + + for ci in range(num_chunks): + windows = chunk_windows[ci] + if not windows: + continue + chunk_start = ci * ttt_chunk + chunk_end = min((ci + 1) * ttt_chunk, total_tokens) + + # --- Phase 1: SCORE this chunk's windows (inference_mode) --- + my_s = (len(windows) * rank) // world_size + my_e = (len(windows) * (rank + 1)) // world_size + my_windows = windows[my_s:my_e] + + base_model.eval() + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk_tok = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk_tok[:-1] + y_batch[i, :wlen] = chunk_tok[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = base_model.forward_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt, prev = y_batch[i, s:wlen], x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + + # --- Phase 2: TRAIN on this chunk (already scored = legal) --- + is_last_chunk = (ci == num_chunks - 1) + if not is_last_chunk and args.ttt_epochs > 0: + base_model.train() + chunk_seqs = (chunk_end - chunk_start) // seq_len + if chunk_seqs > 0: + cos_lr = args.ttt_lr * 0.5 * (1.0 + math.cos(math.pi * ci / max(num_chunks - 1, 1))) + for pg in optimizer.param_groups: + pg['lr'] = cos_lr + my_seq_s = (chunk_seqs * rank) // world_size + my_seq_e = (chunk_seqs * (rank + 1)) // world_size + my_chunk_seqs = my_seq_e - my_seq_s + for _ep in range(args.ttt_epochs): + for bs in range(0, my_chunk_seqs, args.ttt_batch_seqs): + be = min(bs + args.ttt_batch_seqs, my_chunk_seqs) + actual_bs = my_seq_s + bs + start_tok = chunk_start + actual_bs * seq_len + end_tok = chunk_start + (my_seq_s + be) * seq_len + 1 + if end_tok > val_tokens.numel(): + continue + local = val_tokens[start_tok:end_tok].to(device=device, dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + optimizer.zero_grad(set_to_none=True) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + loss = base_model(x, y) + loss.backward() + if world_size > 1: + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + torch.nn.utils.clip_grad_norm_(ttt_params, args.ttt_grad_clip) + optimizer.step() + + if rank == 0 and (ci % 10 == 0 or ci == num_chunks - 1): + elapsed = time.perf_counter() - t0 + rl = loss_sum.item() / max(token_count.item(), 1) + rbpb = rl / math.log(2.0) * (token_count.item() / max(byte_count.item(), 1)) if token_count.item() > 0 else 0.0 + log0(f" ttt_chunk [{ci+1}/{num_chunks}] bpb={rbpb:.6f} time={elapsed:.1f}s") + + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_count.item()) + + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + + log0(f"ttt_sliding:done val_loss={val_loss:.6f} val_bpb={val_bpb:.6f} " + f"elapsed={time.perf_counter() - t0:.1f}s") + return val_loss, val_bpb + + +# --- GPTQ-lite int6 quantization --- + +def _classify_param(name: str) -> str: + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or (".proj." in name and ".mlp." not in name): + return "attn" + return "other" +def quantize_int6_per_row(t: Tensor, clip_range: int = 31) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + best_q, best_s, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(t32.abs(), pct, dim=1) + else: + row_clip = t32.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + recon = q.float() * s.float()[:, None] + err = (t32 - recon).pow(2).mean().item() + if err < best_err: + best_q, best_s, best_err = q, s, err + return best_q, best_s + amax = t32.abs().max().item() + scale = torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + q = torch.clamp(torch.round(t32 / scale.float()), -clip_range, clip_range).to(torch.int8) + return q, scale + +def _unbank_state_dict(sd: dict[str, Tensor], num_layers: int) -> dict[str, Tensor]: + """Convert 3D bank tensors into individual 2D tensors with standard names.""" + out: dict[str, Tensor] = {} + n = num_layers + for name, tensor in sd.items(): + if name == "qo_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_q.weight"] = tensor[i] + out[f"blocks.{i}.attn.proj.weight"] = tensor[n + i] + elif name == "kv_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_k.weight"] = tensor[i] + out[f"blocks.{i}.attn.c_v.weight"] = tensor[n + i] + elif name == "mlp_up_bank": + for i in range(n): + out[f"blocks.{i}.mlp.fc.weight"] = tensor[i] + elif name == "mlp_down_bank": + for i in range(n): + out[f"blocks.{i}.mlp.proj.weight"] = tensor[i] + else: + out[name] = tensor + return out + +def _rebank_state_dict(sd: dict[str, Tensor], num_layers: int, template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + """Convert individual 2D tensors back into 3D bank tensors.""" + out: dict[str, Tensor] = {} + n = num_layers + # Reconstruct banks from individual weight keys + qo_slices = [None] * (2 * n) + kv_slices = [None] * (2 * n) + up_slices = [None] * n + down_slices = [None] * n + consumed = set() + for i in range(n): + qk = f"blocks.{i}.attn.c_q.weight" + if qk in sd: + qo_slices[i] = sd[qk] + consumed.add(qk) + ok = f"blocks.{i}.attn.proj.weight" + if ok in sd: + qo_slices[n + i] = sd[ok] + consumed.add(ok) + kk = f"blocks.{i}.attn.c_k.weight" + if kk in sd: + kv_slices[i] = sd[kk] + consumed.add(kk) + vk = f"blocks.{i}.attn.c_v.weight" + if vk in sd: + kv_slices[n + i] = sd[vk] + consumed.add(vk) + fk = f"blocks.{i}.mlp.fc.weight" + if fk in sd: + up_slices[i] = sd[fk] + consumed.add(fk) + dk = f"blocks.{i}.mlp.proj.weight" + if dk in sd: + down_slices[i] = sd[dk] + consumed.add(dk) + out["qo_bank"] = torch.stack(qo_slices).to(dtype=template_sd["qo_bank"].dtype) + out["kv_bank"] = torch.stack(kv_slices).to(dtype=template_sd["kv_bank"].dtype) + out["mlp_up_bank"] = torch.stack(up_slices).to(dtype=template_sd["mlp_up_bank"].dtype) + out["mlp_down_bank"] = torch.stack(down_slices).to(dtype=template_sd["mlp_down_bank"].dtype) + for name, tensor in sd.items(): + if name not in consumed: + out[name] = tensor + return out + +def mixed_quantize_int6(state_dict: dict[str, Tensor], int6_cats: set[str], clip_range: int = 31, + hessians: dict[str, Tensor] | None = None): + num_layers_total = max( + (int(k.split(".")[1]) for k in state_dict if k.startswith("blocks.")), + default=0, + ) + 1 + late_k_layers = set(range(num_layers_total - 2, num_layers_total)) + result: dict[str, Tensor] = {} + meta: dict[str, object] = {} + gptq_count, naive_count = 0, 0 + for name, tensor in state_dict.items(): + t = tensor.detach().cpu().contiguous() + cat = _classify_param(name) + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough" + continue + if any(p in name for p in CONTROL_TENSOR_NAME_PATTERNS): + result[name] = t.float() + meta[name] = "passthrough_ctrl" + continue + if cat in int6_cats and t.ndim >= 1: + H = hessians.get(name) if hessians else None + if H is not None and t.ndim == 2: + q, s = gptq_quantize_weight(t, H.cpu(), clip_range=clip_range) + gptq_count += 1 + else: + q, s = quantize_int6_per_row(t, clip_range=clip_range) + naive_count += 1 + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int6"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + if hessians: + print(f"gptq_quantize: {gptq_count} GPTQ layers, {naive_count} naive layers", flush=True) + return result, meta +def dequantize_mixed_int6(result: dict[str, Tensor], meta: dict[str, object], + template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + for name, orig in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if info in ("passthrough", "passthrough_ctrl", "passthrough_fp16"): + t = result[name] + if t.dtype == torch.float16 and orig_dtype in (torch.float32, torch.bfloat16): + t = t.to(orig_dtype) + out[name] = t + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out + +# --- Full Hessian GPTQ --- + +def gptq_quantize_weight(W: Tensor, H: Tensor, clip_range: int = 31, + block_size: int = 128, percdamp: float = 0.01) -> tuple[Tensor, Tensor]: + """GPTQ with Cholesky error compensation and actorder (Frantar et al., ICLR 2023).""" + W_orig = W.float().clone() + rows, cols = W_orig.shape + H = H.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = percdamp * H.diag().mean() + H.diagonal().add_(damp) + perm = torch.argsort(H.diag(), descending=True) + invperm = torch.argsort(perm) + W_perm = W_orig[:, perm].clone() + W_perm[:, dead[perm]] = 0 + H = H[perm][:, perm] + try: + Hinv = torch.cholesky_inverse(torch.linalg.cholesky(H)) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + except torch.linalg.LinAlgError: + return quantize_int6_per_row(W_orig, clip_range) + best_q, best_scale, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(W_orig.abs(), pct, dim=1) + else: + row_clip = W_orig.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + sf = s.float() + Q = torch.zeros(rows, cols, dtype=torch.int8) + W_work = W_perm.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + W_block = W_work[:, i1:i2].clone() + Hinv_block = Hinv[i1:i2, i1:i2] + Err = torch.zeros(rows, i2 - i1) + for j in range(i2 - i1): + w_col = W_block[:, j] + d = Hinv_block[j, j] + q_col = torch.clamp(torch.round(w_col / sf), -clip_range, clip_range) + Q[:, i1 + j] = q_col.to(torch.int8) + err = (w_col - q_col.float() * sf) / d + Err[:, j] = err + W_block[:, j:] -= err.unsqueeze(1) * Hinv_block[j, j:].unsqueeze(0) + if i2 < cols: + W_work[:, i2:] -= Err @ Hinv[i1:i2, i2:] + recon = Q.float() * sf[:, None] + mse = (W_perm - recon).pow(2).mean().item() + if mse < best_err: + best_q, best_scale, best_err = Q, s, mse + best_q = best_q[:, invperm] + return best_q, best_scale + +def _init_hessians(nl: int, dim: int, mlp_dim: int, device: torch.device) -> dict[str, Tensor]: + h: dict[str, Tensor] = {} + for i in range(nl): + for k in ['c_q', 'c_k', 'c_v']: + h[f'blocks.{i}.attn.{k}.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.attn.proj.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.mlp.fc.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.mlp.proj.weight'] = torch.zeros(mlp_dim, mlp_dim, dtype=torch.float32, device=device) + return h + +def _accum_hessians(hessians: dict[str, Tensor], blocks: nn.ModuleList, dim: int, mlp_dim: int) -> None: + for i, block in enumerate(blocks): + qkv_in = block.attn._gptq_qkv_in.float().reshape(-1, dim) + h_qkv = qkv_in.t() @ qkv_in + hessians[f'blocks.{i}.attn.c_q.weight'] += h_qkv + hessians[f'blocks.{i}.attn.c_k.weight'] += h_qkv + hessians[f'blocks.{i}.attn.c_v.weight'] += h_qkv + o_in = block.attn._gptq_o_in.float().reshape(-1, dim) + hessians[f'blocks.{i}.attn.proj.weight'] += o_in.t() @ o_in + up_in = block.mlp._gptq_up_in.float().reshape(-1, dim) + hessians[f'blocks.{i}.mlp.fc.weight'] += up_in.t() @ up_in + down_in = block.mlp._gptq_down_in.float().reshape(-1, mlp_dim) + hessians[f'blocks.{i}.mlp.proj.weight'] += down_in.t() @ down_in + +def _finalize_hessians(hessians: dict[str, Tensor], num_batches: int) -> None: + for name in hessians: + hessians[name] = hessians[name].cpu() / num_batches + damp = 0.01 * torch.diag(hessians[name]).mean().clamp_min(1e-6) + hessians[name] += damp * torch.eye(hessians[name].shape[0]) + +def gptq_collect_hessians(base_model: nn.Module, train_loader, device: torch.device, + num_batches: int, batch_tokens: int, seq_len: int, + grad_accum_steps: int) -> dict[str, Tensor]: + """Collect Hessians H = X^T X from training data.""" + nl = base_model.num_layers + dim = base_model.tok_emb.weight.shape[1] + mlp_dim = base_model.mlp_up_bank.shape[1] + hessians = _init_hessians(nl, dim, mlp_dim, device) + for block in base_model.blocks: + block.attn._save_gptq = True + block.mlp._save_gptq = True + base_model.eval() + with torch.inference_mode(), torch.autocast(device_type='cuda', dtype=torch.bfloat16): + for _ in range(num_batches): + x, y = train_loader.next_batch(batch_tokens, seq_len, grad_accum_steps) + base_model(x, y) + _accum_hessians(hessians, base_model.blocks, dim, mlp_dim) + for block in base_model.blocks: + block.attn._save_gptq = False + block.mlp._save_gptq = False + _finalize_hessians(hessians, num_batches) + base_model.train() + return hessians + +# --- Training --- + +def main() -> None: + code = Path(__file__).read_text(encoding="utf-8") + args = Hyperparameters() + # zeropower_via_newtonschulz5 runs eagerly with bmm -- do NOT compile + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError(f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral") + grad_accum_steps = 8 // world_size + grad_scale = 1.0 / grad_accum_steps + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + master_process = rank == 0 + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + logfile = None + if master_process: + os.makedirs("logs", exist_ok=True) + logfile = f"logs/{args.run_id}.txt" + print(logfile) + def log0(msg: str, console: bool = True) -> None: + if not master_process: + return + if console: + print(msg) + if logfile is not None: + with open(logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + log0(code, console=False) + log0("=" * 100, console=False) + log0(f"Running Python {sys.version}", console=False) + log0(f"Running PyTorch {torch.__version__}", console=False) + log0( + subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=False).stdout, + console=False, + ) + log0("=" * 100, console=False) + random.seed(args.seed) + np.random.seed(args.seed) + torch.manual_seed(args.seed) + torch.cuda.manual_seed_all(args.seed) + (base_bytes_lut, has_leading_space_lut, is_boundary_token_lut), tokenizer_metadata = load_tokenizer_luts( + args.tokenizer_path, args.tokenizer_meta_path, args.vocab_size, device, + validate_meta=args.tokenizer_meta_validate, + ) + log0(f"tokenizer: kind={tokenizer_metadata.get('tokenizer_kind', 'unknown')} vocab={tokenizer_metadata.get('vocab_size', '?')}") + if tokenizer_metadata.get('tokenizer_kind') == 'sentencepiece': + sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) + if int(sp.vocab_size()) != args.vocab_size: + raise ValueError( + f"VOCAB_SIZE={args.vocab_size} does not match tokenizer vocab_size={int(sp.vocab_size())}" + ) + dataset_dir = Path(args.data_path).resolve() + actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) + effective_eval_seq_len = args.eval_seq_len if args.eval_seq_len > 0 else args.train_seq_len + val_seq_len = max(args.train_seq_len, effective_eval_seq_len) + val_tokens = load_validation_tokens(args.val_files, val_seq_len) + log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") + log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") + base_model = GPT( + vocab_size=args.vocab_size, + num_layers=args.num_layers, + model_dim=args.model_dim, + num_heads=args.num_heads, + num_kv_heads=args.num_kv_heads, + mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, + tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, + rope_base=args.rope_base, + qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, + bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, + ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, + ve_dim=args.ve_dim, + ve_layers=args.ve_layers, + neg_slope=args.negative_slope, + ).to(device).bfloat16() + # Banks stay FP32 (like CastedLinear weights), cast to BF16 in forward + base_model.qo_bank.data = base_model.qo_bank.data.float() + base_model.kv_bank.data = base_model.kv_bank.data.float() + base_model.mlp_up_bank.data = base_model.mlp_up_bank.data.float() + base_model.mlp_down_bank.data = base_model.mlp_down_bank.data.float() + for module in base_model.modules(): + if isinstance(module, CastedLinear): + module.float() + restore_low_dim_params_to_fp32(base_model) + # No DDP -- Parallel Muon handles bank grad communication via reduce-scatter, + # and non-bank grads are manually all-reduced before Adam steps. + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + model = compiled_model + + # Optimizer split: + # - 4 parameter banks -> Muon (batched Newton-Schulz) + # - token embedding -> Adam + # - scalars/control tensors -> Adam + # - bigram proj, VE proj -> Adam (small matrix params not worth banking) + matrix_params = [ + base_model.qo_bank, base_model.kv_bank, + base_model.mlp_up_bank, base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for name, p in block_named_params + if p.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + scalar_params.append(base_model.smear.gate) + if base_model.bigram is not None: + scalar_params.append(base_model.bigram.scale) + token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr + tok_params = [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}] + if base_model.bigram is not None: + tok_params.append({"params": [base_model.bigram.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.bigram.proj is not None: + scalar_params.append(base_model.bigram.proj.weight) + if base_model.ve_shared is not None: + tok_params.append({"params": [base_model.ve_shared.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.ve_shared.proj is not None: + scalar_params.append(base_model.ve_shared.proj.weight) + scalar_params.append(base_model.ve_shared.scale) + for s in base_model.ve_layer_scales: + scalar_params.append(s) + optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + optimizer_muon = Muon( + matrix_params, + lr=args.matrix_lr, + momentum=args.muon_momentum, + backend_steps=args.muon_backend_steps, + weight_decay=args.muon_wd, + ) + for group in optimizer_muon.param_groups: + group["base_lr"] = args.matrix_lr + optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + # Non-bank params that need manual all-reduce (replicated across GPUs) + replicated_params = list(optimizer_tok.param_groups[0]["params"]) + for pg in optimizer_tok.param_groups[1:]: + replicated_params.extend(pg["params"]) + replicated_params.extend(scalar_params) + + optimizer_head = None + if base_model.lm_head is not None: + optimizer_head = torch.optim.Adam( + [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + replicated_params.append(base_model.lm_head.weight) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if optimizer_head is not None: + optimizers.append(optimizer_head) + log0(f"model_params:{sum(p.numel() for p in base_model.parameters())}") + xsa_layers = [i for i, b in enumerate(base_model.blocks) if b.attn.use_xsa] + log0(f"XSA:last_{args.xsa_last_n} active_layers:{xsa_layers}") + log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") + log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") + log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") + log0( + f"tie_embeddings:{args.tie_embeddings} embed_lr:{token_lr} " + f"head_lr:{args.head_lr if base_model.lm_head is not None else 0.0} " + f"matrix_lr:{args.matrix_lr} scalar_lr:{args.scalar_lr}" + ) + log0( + f"train_batch_tokens:{args.train_batch_tokens} train_seq_len:{args.train_seq_len} " + f"iterations:{args.iterations} warmup_steps:{args.warmup_steps} " + f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" + ) + log0(f"seed:{args.seed}") + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + def zero_grad_all() -> None: + for opt in optimizers: + opt.zero_grad(set_to_none=True) + max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None + if args.use_gptq and max_wallclock_ms is not None: + max_wallclock_ms -= args.gptq_reserve_ms + log0(f"gptq:reserving {args.gptq_reserve_ms:.0f}ms from training budget, effective={max_wallclock_ms:.0f}ms") + def lr_mul(step: int, elapsed_ms: float) -> float: + if args.warmdown_iters <= 0: + return 1.0 + if max_wallclock_ms is None: + warmdown_start = max(args.iterations - args.warmdown_iters, 0) + return max((args.iterations - step) / max(args.warmdown_iters, 1), 0.0) if warmdown_start <= step < args.iterations else 1.0 + step_ms = elapsed_ms / max(step, 1) + warmdown_ms = args.warmdown_iters * step_ms + remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) + return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 + if args.warmup_steps > 0: + initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} + initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] + model.train() + for warmup_step in range(args.warmup_steps): + zero_grad_all() + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + warmup_loss = model(x, y) + (warmup_loss * grad_scale).backward() + # All-reduce all grads for warmup (simple, not optimized) + if distributed: + for p in base_model.parameters(): + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + for opt in optimizers: + opt.step() + zero_grad_all() + if args.warmup_steps <= 20 or (warmup_step + 1) % 10 == 0 or warmup_step + 1 == args.warmup_steps: + log0(f"warmup_step:{warmup_step + 1}/{args.warmup_steps}") + base_model.load_state_dict(initial_model_state, strict=True) + for opt, state in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + zero_grad_all() + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + swa_state: dict[str, Tensor] | None = None + swa_count = 0 + ema_state = {name: t.detach().float().clone() for name, t in base_model.state_dict().items()} + ema_decay = 0.997 + training_time_ms = 0.0 + stop_after_step: int | None = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) + should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1000.0 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + log0( + f"step:{step}/{args.iterations} val_loss:{val_loss:.4f} val_bpb:{val_bpb:.4f} " + f"train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms / max(step, 1):.2f}ms" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < args.iterations: + log0( + f"stopping_early: wallclock_cap train_time:{training_time_ms:.0f}ms " + f"step:{step}/{args.iterations}" + ) + break + elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + scale = lr_mul(step, elapsed_ms) + zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y) + train_loss += loss.detach() + (loss * grad_scale).backward() + train_loss /= grad_accum_steps + frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 + muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum + for group in optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * scale + if args.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) + # === 3-phase overlapped optimizer step === + # Phase 1: Launch async reduce-scatter for banks (biggest first) + optimizer_muon.launch_reduce_scatters() + # Phase 2: All-reduce non-bank grads + step Adam (while bank RS is in-flight) + if distributed: + for p in replicated_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + optimizer_tok.step() + optimizer_scalar.step() + if optimizer_head is not None: + optimizer_head.step() + # Phase 3: Wait for RS, local NS5, all-gather (banks processed last) + optimizer_muon.step() + zero_grad_all() + # EMA update + with torch.no_grad(): + for name, t in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_(t.detach().float(), alpha=1.0 - ema_decay) + step += 1 + approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + if args.swa_enabled and scale < 0.2 and step % args.swa_every == 0: + if swa_state is None: + swa_state = {name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()} + swa_count = 1 + log0(f"swa:start step:{step}") + else: + for name, t in base_model.state_dict().items(): + swa_state[name] += t.detach().cpu() + swa_count += 1 + should_log_train = ( + args.train_log_every > 0 + and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) + ) + if should_log_train: + log0( + f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " + f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" + ) + reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + if distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log0( + f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " + f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" + ) + # Apply EMA weights + log0("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = {name: t.to(dtype=current_state[name].dtype) for name, t in ema_state.items()} + base_model.load_state_dict(avg_state, strict=True) + torch.cuda.synchronize() + t_diag = time.perf_counter() + diag_val_loss, diag_val_bpb = eval_val( + args, compiled_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"DIAGNOSTIC post_ema val_loss:{diag_val_loss:.4f} val_bpb:{diag_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_diag):.0f}ms" + ) + export_sd = base_model.state_dict() + if master_process: + torch.save(export_sd, "final_model.pt") + model_bytes = os.path.getsize("final_model.pt") + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model: {model_bytes} bytes") + log0(f"Code size: {code_bytes} bytes") + # Unbank 3D tensors into individual 2D tensors for quantization + sd_cpu = {k: v.detach().cpu() for k, v in export_sd.items()} + unbanked_sd = _unbank_state_dict(sd_cpu, args.num_layers) + # GPTQ calibration: collect Hessians from training data + gptq_hessians = None + if args.use_gptq: + t_gptq = time.perf_counter() + log0(f"gptq:calibrating with {args.gptq_calib_samples} batches (training data)...") + calib_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + gptq_hessians = gptq_collect_hessians( + base_model, calib_loader, device, num_batches=args.gptq_calib_samples, + batch_tokens=args.train_batch_tokens, seq_len=args.train_seq_len, + grad_accum_steps=grad_accum_steps) + del calib_loader + gptq_elapsed = time.perf_counter() - t_gptq + log0(f"gptq:calibrated {len(gptq_hessians)} layers in {gptq_elapsed:.1f}s") + torch.cuda.empty_cache() + quant_result, quant_meta = mixed_quantize_int6(unbanked_sd, {"mlp", "attn"}, clip_range=args.quant_clip_range, hessians=gptq_hessians) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = lzma.compress(quant_raw, preset=6) + if master_process: + with open("final_model.int6.ptz", "wb") as f: + f.write(quant_blob) + quant_file_bytes = len(quant_blob) + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model int6+lzma: {quant_file_bytes} bytes") + log0(f"Total submission size int6+lzma: {quant_file_bytes + code_bytes} bytes") + if distributed: + dist.barrier() + with open("final_model.int6.ptz", "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(lzma.decompress(quant_blob_disk)), + map_location="cpu", + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd) + # Re-bank the dequantized tensors + deq_state = _rebank_state_dict(deq_unbanked, args.num_layers, sd_cpu) + eval_model = GPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + neg_slope=args.negative_slope, + ).to(device).bfloat16() + eval_model.qo_bank.data = eval_model.qo_bank.data.float() + eval_model.kv_bank.data = eval_model.kv_bank.data.float() + eval_model.mlp_up_bank.data = eval_model.mlp_up_bank.data.float() + eval_model.mlp_down_bank.data = eval_model.mlp_down_bank.data.float() + for m in eval_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(eval_model) + eval_model.load_state_dict(deq_state, strict=True) + compiled_eval = torch.compile(eval_model, dynamic=False, fullgraph=True) + torch.cuda.synchronize() + t_qeval = time.perf_counter() + q_val_loss, q_val_bpb = eval_val( + args, compiled_eval, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + eval_seq_len=effective_eval_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" + ) + log0(f"final_int6_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + sw_seq_len = effective_eval_seq_len + if args.eval_stride > 0 and args.eval_stride < sw_seq_len: + torch.cuda.synchronize() + t_slide = time.perf_counter() + sw_val_loss, sw_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window val_loss:{sw_val_loss:.4f} val_bpb:{sw_val_bpb:.4f} " + f"stride:{args.eval_stride} eval_time:{1000.0 * (time.perf_counter() - t_slide):.0f}ms" + ) + log0(f"final_int6_sliding_window_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + if args.eval_stride != 64 and 64 < sw_seq_len: + torch.cuda.synchronize() + t_slide64 = time.perf_counter() + sw64_val_loss, sw64_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=64, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window_s64 val_loss:{sw64_val_loss:.4f} val_bpb:{sw64_val_bpb:.4f} " + f"stride:64 eval_time:{1000.0 * (time.perf_counter() - t_slide64):.0f}ms" + ) + log0(f"final_int6_sliding_window_s64_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + # Legal score-first TTT (PR #461 recipe) + if args.ttt_enabled: + torch.cuda.synchronize() + t_ttt = time.perf_counter() + ttt_loss, ttt_bpb = eval_val_sliding_ttt( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, log0=log0, + ) + torch.cuda.synchronize() + log0(f"legal_ttt val_loss:{ttt_loss:.4f} val_bpb:{ttt_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt):.0f}ms") + log0(f"legal_ttt_exact val_loss:{ttt_loss:.8f} val_bpb:{ttt_bpb:.8f}") + if distributed: + dist.destroy_process_group() +if __name__ == "__main__": + main() diff --git a/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed1337.log b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed1337.log new file mode 100644 index 0000000000..0e6403215a --- /dev/null +++ b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed1337.log @@ -0,0 +1,2505 @@ +from __future__ import annotations +import copy +import glob +import io +import lzma +import math +import os +import random +import subprocess +import sys +import time +import uuid +from pathlib import Path +import numpy as np +import sentencepiece as spm +import torch +import torch.distributed as dist +import torch.nn.functional as F +from torch import Tensor, nn +from torch.nn.parallel import DistributedDataParallel as DDP +from flash_attn_interface import flash_attn_func as flash_attn_3_func +class Hyperparameters: + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp1024") + train_files = os.path.join(data_path, "fineweb_train_*.bin") + val_files = os.path.join(data_path, "fineweb_val_*.bin") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_1024_bpe.model") + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + seed = int(os.environ.get("SEED", 1337)) + val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 3500)) + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786_432)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 1.5)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 1024)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 3.0)) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.035)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.025)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.025)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.99)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + swa_enabled = bool(int(os.environ.get("SWA_ENABLED", "1"))) + swa_every = int(os.environ.get("SWA_EVERY", 50)) + muon_wd = float(os.environ.get("MUON_WD", 0.04)) + adam_wd = float(os.environ.get("ADAM_WD", 0.04)) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 2048)) + bigram_dim = int(os.environ.get("BIGRAM_DIM", 128)) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 4)) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + ve_enabled = bool(int(os.environ.get("VE_ENABLED", "1"))) + ve_dim = int(os.environ.get("VE_DIM", 128)) + ve_layers = os.environ.get("VE_LAYERS", "9,10") + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "0"))) + ttt_lr = float(os.environ.get("TTT_LR", 0.002)) + ttt_epochs = int(os.environ.get("TTT_EPOCHS", 3)) + ttt_chunk_tokens = int(os.environ.get("TTT_CHUNK_TOKENS", 32768)) + ttt_freeze_blocks = int(os.environ.get("TTT_FREEZE_BLOCKS", 2)) + ttt_momentum = float(os.environ.get("TTT_MOMENTUM", 0.9)) + ttt_batch_seqs = int(os.environ.get("TTT_BATCH_SEQS", 32)) + ttt_grad_clip = float(os.environ.get("TTT_GRAD_CLIP", 1.0)) + negative_slope = float(os.environ.get("NEGATIVE_SLOPE", 0.5)) + use_gptq = bool(int(os.environ.get("USE_GPTQ", "0"))) + gptq_calib_samples = int(os.environ.get("GPTQ_CALIB_SAMPLES", "64")) + gptq_reserve_ms = float(os.environ.get("GPTQ_RESERVE_MS", "14000")) + quant_clip_range = int(os.environ.get("QUANT_CLIP_RANGE", 31)) + tokenizer_meta_path = os.environ.get("TOKENIZER_META_PATH", "") + tokenizer_meta_validate = bool(int(os.environ.get("TOKENIZER_META_VALIDATE", "0"))) + +# --- Batched Newton-Schulz orthogonalization --- + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 5, eps: float = 1e-7) -> Tensor: + """Batched Newton-Schulz orthogonalization. G: (B,M,N) or (M,N).""" + a, b, c = (3.4445, -4.7750, 2.0315) + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + for _ in range(steps): + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + +# --- Parallel Muon optimizer --- + +class Muon(torch.optim.Optimizer): + """Parallel Muon: post-backward reduce-scatter -> local NS5 -> all-gather. + + No DDP for bank params. After backward, this optimizer: + 1. Launches async reduce-scatter for all banks (biggest first) + 2. Returns control so Adam can step on small params while RS is in-flight + 3. Waits for each RS, runs local NS5 on the shard, launches async all-gather + 4. Each all-gather overlaps with next bank's NS5 + """ + def __init__(self, params, lr: float, momentum: float, backend_steps: int, + nesterov: bool = True, weight_decay: float = 0.0): + super().__init__( + params, + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, + nesterov=nesterov, weight_decay=weight_decay), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + 'p': p, + 'B': B, + 'padded_grad': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard_mom': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'full_update': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'scale': max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + # Sort by size descending -- launch biggest reduce-scatters first + self._bank_meta.sort(key=lambda m: -m['p'].numel()) + self._built = True + + def launch_reduce_scatters(self): + """Phase 1: launch async reduce-scatter for all banks. Call right after backward.""" + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m['p'] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m['padded_grad'] + pg[:m['B']].copy_(p.grad.bfloat16()) + if pg.shape[0] > m['B']: + pg[m['B']:].zero_() + fut = dist.reduce_scatter_tensor(m['shard'], pg, op=dist.ReduceOp.AVG, async_op=True) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + """Phase 3: wait for RS, local NS5, all-gather. Call AFTER Adam steps.""" + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + + if not self._built: + self._build() + + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + + prev_ag_handle = None + prev_m = None + + sharded = self._distributed and hasattr(self, '_rs_futures') + + for i, m in enumerate(self._bank_meta): + p = m['p'] + if p.grad is None: + continue + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if sharded and self._rs_futures[i] is not None: + self._rs_futures[i].wait() + g = m['shard'] + buf = m['shard_mom'] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m['full_update'], update, async_op=True) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m['scale']) + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if hasattr(self, '_rs_futures'): + del self._rs_futures + + return loss + +# --- Tokenizer evaluation helpers --- + +def build_sentencepiece_luts( + sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device +) -> tuple[Tensor, Tensor, Tensor]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) + +TOKENIZER_META_FORMAT_VERSION = 1 +TOKENIZER_META_SUFFIX = ".meta.npz" + + +def _derive_tokenizer_meta_path(tokenizer_path: str) -> Path: + tokenizer = Path(tokenizer_path) + if tokenizer.suffix == ".model": + return tokenizer.with_suffix(TOKENIZER_META_SUFFIX) + return tokenizer.with_name(tokenizer.name + TOKENIZER_META_SUFFIX) + + +def build_sentencepiece_luts_np( + sp: spm.SentencePieceProcessor, vocab_size: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return base_bytes_np, has_leading_space_np, is_boundary_token_np + + +def load_tokenizer_meta_luts_np( + meta_path: Path, vocab_size: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray, dict[str, object]]: + def _scalar(value): + arr = np.asarray(value) + if arr.ndim == 0: + return arr.item() + first = arr.reshape(-1)[0] + return first.item() if hasattr(first, "item") else first + + with np.load(meta_path, allow_pickle=False) as data: + format_version = int(_scalar(data["format_version"])) + if format_version != TOKENIZER_META_FORMAT_VERSION: + raise ValueError( + f"Unsupported tokenizer meta format_version={format_version} " + f"expected={TOKENIZER_META_FORMAT_VERSION}" + ) + meta_vocab_size = int(_scalar(data["vocab_size"])) + tokenizer_kind = str(_scalar(data["tokenizer_kind"])) + source_model_name = str(_scalar(data["source_model_name"])) + base_bytes_np = np.asarray(data["base_bytes"], dtype=np.int16) + has_leading_space_np = np.asarray(data["has_leading_space"], dtype=np.bool_) + is_boundary_token_np = np.asarray(data["is_boundary_token"], dtype=np.bool_) + table_size = max(meta_vocab_size, vocab_size) + if base_bytes_np.shape[0] < table_size: + padded_base_bytes = np.zeros((table_size,), dtype=np.int16) + padded_has_leading_space = np.zeros((table_size,), dtype=np.bool_) + padded_is_boundary = np.ones((table_size,), dtype=np.bool_) + padded_base_bytes[: base_bytes_np.shape[0]] = base_bytes_np + padded_has_leading_space[: has_leading_space_np.shape[0]] = has_leading_space_np + padded_is_boundary[: is_boundary_token_np.shape[0]] = is_boundary_token_np + base_bytes_np = padded_base_bytes + has_leading_space_np = padded_has_leading_space + is_boundary_token_np = padded_is_boundary + metadata = { + "format_version": format_version, + "tokenizer_kind": tokenizer_kind, + "source_model_name": source_model_name, + "vocab_size": meta_vocab_size, + "meta_path": str(meta_path), + } + return base_bytes_np, has_leading_space_np, is_boundary_token_np, metadata + + +def load_tokenizer_luts( + tokenizer_path: str, + tokenizer_meta_path: str, + vocab_size: int, + device: torch.device, + *, + validate_meta: bool = False, +) -> tuple[tuple[Tensor, Tensor, Tensor], dict[str, object]]: + meta_path = ( + Path(tokenizer_meta_path) if tokenizer_meta_path + else _derive_tokenizer_meta_path(tokenizer_path) + ) + if meta_path.exists(): + base_bytes_np, has_leading_space_np, is_boundary_token_np, metadata = ( + load_tokenizer_meta_luts_np(meta_path, vocab_size) + ) + if validate_meta and str(tokenizer_path).endswith(".model"): + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + sp_luts = build_sentencepiece_luts_np(sp, vocab_size) + if not ( + np.array_equal(base_bytes_np, sp_luts[0]) + and np.array_equal(has_leading_space_np, sp_luts[1]) + and np.array_equal(is_boundary_token_np, sp_luts[2]) + ): + raise ValueError(f"Tokenizer metadata mismatch for {meta_path}") + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ), metadata + if not str(tokenizer_path).endswith(".model"): + raise FileNotFoundError(f"TOKENIZER_META_PATH does not exist: {meta_path}") + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + return build_sentencepiece_luts(sp, vocab_size, device), { + "tokenizer_kind": "sentencepiece", + "source_model_name": str(tokenizer_path), + "vocab_size": int(sp.vocab_size()), + "meta_path": None, + "fallback": True, + } + +def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = ((tokens.numel() - 1) // seq_len) * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] +def eval_val( + args: Hyperparameters, + model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + grad_accum_steps: int, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + seq_len = eval_seq_len or args.train_seq_len + local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + "VAL_BATCH_SIZE must provide at least one sequence per rank; " + f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " + f"GRAD_ACCUM_STEPS={grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + seq_start = (total_seqs * rank) // world_size + seq_end = (total_seqs * (rank + 1)) // world_size + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + model.eval() + with torch.inference_mode(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + batch_loss = model(x, y).detach() + batch_token_count = float(y.numel()) + val_loss_sum += batch_loss.to(torch.float64) * batch_token_count + val_token_count += batch_token_count + prev_ids = x.reshape(-1) + tgt_ids = y.reshape(-1) + token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + val_loss = val_loss_sum / val_token_count + bits_per_token = val_loss.item() / math.log(2.0) + tokens_per_byte = val_token_count.item() / val_byte_count.item() + model.train() + return float(val_loss.item()), float(bits_per_token * tokens_per_byte) + +# --- Quantization helpers --- + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,smear,dtg_gate,ve_layer_scales,ve_shared.scale,attn_gate,vr_lambda", + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_FP32_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "INT8_KEEP_FLOAT_FP32_NAME_PATTERNS", + ",".join(CONTROL_TENSOR_NAME_PATTERNS), + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_MAX_NUMEL = 65_536 +INT8_KEEP_FLOAT_STORE_DTYPE = torch.float16 +INT8_PER_ROW_SCALE_DTYPE = torch.float16 +INT8_CLIP_PERCENTILE = 99.99984 +INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 +def tensor_nbytes(t: Tensor) -> int: + return int(t.numel()) * int(t.element_size()) +def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: + if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): + return t.float().contiguous() + if t.dtype in {torch.float32, torch.bfloat16}: + passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") + return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() + return t +def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + clip_abs = ( + torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) + if t32.numel() + else torch.empty((t32.shape[0],), dtype=torch.float32) + ) + clipped = torch.maximum(torch.minimum(t32, clip_abs[:, None]), -clip_abs[:, None]) + scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) + q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() + return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() + clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 + scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) + q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() + return q, scale +def quantize_state_dict_int8(state_dict: dict[str, Tensor]): + quantized: dict[str, Tensor] = {} + scales: dict[str, Tensor] = {} + dtypes: dict[str, str] = {} + passthrough: dict[str, Tensor] = {} + passthrough_orig_dtypes: dict[str, str] = {} + qmeta: dict[str, dict[str, object]] = {} + stats = dict.fromkeys( + ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), + 0, + ) + for name, tensor in state_dict.items(): + t = tensor.detach().to("cpu").contiguous() + stats["param_count"] += int(t.numel()) + stats["num_tensors"] += 1 + stats["baseline_tensor_bytes"] += tensor_nbytes(t) + if not t.is_floating_point(): + stats["num_nonfloat_tensors"] += 1 + passthrough[name] = t + stats["int8_payload_bytes"] += tensor_nbytes(t) + continue + if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: + kept = keep_float_tensor(name, t, passthrough_orig_dtypes) + passthrough[name] = kept + stats["int8_payload_bytes"] += tensor_nbytes(kept) + continue + stats["num_float_tensors"] += 1 + q, s = quantize_float_tensor(t) + if s.ndim > 0: + qmeta[name] = {"scheme": "per_row", "axis": 0} + quantized[name] = q + scales[name] = s + dtypes[name] = str(t.dtype).removeprefix("torch.") + stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) + obj: dict[str, object] = { + "__quant_format__": "int8_clean_per_row_v1", + "quantized": quantized, + "scales": scales, + "dtypes": dtypes, + "passthrough": passthrough, + } + if qmeta: + obj["qmeta"] = qmeta + if passthrough_orig_dtypes: + obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes + return obj, stats +def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + qmeta = obj.get("qmeta", {}) + passthrough_orig_dtypes = obj.get("passthrough_orig_dtypes", {}) + for name, q in obj["quantized"].items(): + dtype = getattr(torch, obj["dtypes"][name]) + s = obj["scales"][name] + if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: + s = s.to(dtype=torch.float32) + out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() + else: + scale = float(s.item()) + out[name] = (q.float() * scale).to(dtype=dtype).contiguous() + for name, t in obj["passthrough"].items(): + out_t = t.detach().to("cpu").contiguous() + orig_dtype = passthrough_orig_dtypes.get(name) + if isinstance(orig_dtype, str): + out_t = out_t.to(dtype=getattr(torch, orig_dtype)).contiguous() + out[name] = out_t + return out + +# --- Data loading --- + +def load_data_shard(file: Path) -> Tensor: + header_bytes = 256 * np.dtype(" int: + key = str(file) + cached = _SHARD_NTOKENS_CACHE.get(key) + if cached is not None: + return cached + header = np.fromfile(file, dtype=" np.memmap: + key = str(file) + mm = _MMAP_CACHE.get(key) + if mm is not None: + return mm + n = _read_num_tokens(file) + mm = np.memmap(file, mode="r", dtype=" int: + if n <= 1: + return 1 + while True: + s = int(self._rng.integers(1, n)) + if math.gcd(s, n) == 1: + return s + def _reset_cursor(self, si: int, seq_len: int) -> None: + nt = int(self._num_tokens[si]) + max_phase = min(seq_len - 1, max(0, nt - seq_len - 1)) + phase = int(self._rng.integers(max_phase + 1)) if max_phase > 0 else 0 + bc = (nt - 1 - phase) // seq_len + self._cursor_phase[si] = phase + self._cursor_block_count[si] = bc + self._cursor_next[si] = 0 + self._cursor_start[si] = int(self._rng.integers(bc)) if bc > 1 else 0 + self._cursor_stride[si] = self._pick_coprime_stride(bc) + self._cursor_init[si] = True + def _ensure_cursor(self, si: int, seq_len: int) -> None: + if not self._cursor_init[si] or self._cursor_next[si] >= self._cursor_block_count[si]: + self._reset_cursor(si, seq_len) + def _take_from_shard(self, si: int, seq_len: int, count: int, out: list[tuple[int, int]]) -> None: + rem = count + while rem > 0: + self._ensure_cursor(si, seq_len) + bc = int(self._cursor_block_count[si]) + ni = int(self._cursor_next[si]) + take = min(rem, bc - ni) + phase = int(self._cursor_phase[si]) + start = int(self._cursor_start[si]) + stride = int(self._cursor_stride[si]) + for j in range(take): + bi = (start + (ni + j) * stride) % bc + out.append((si, phase + bi * seq_len)) + self._cursor_next[si] = ni + take + rem -= take + def _init_pipeline(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> None: + local_tokens = global_tokens // (self.world_size * grad_accum_steps) + num_seqs = local_tokens // seq_len + global_num_seqs = num_seqs * self.world_size + self._cfg = (local_tokens, seq_len, num_seqs, global_num_seqs) + bbc = (self._num_tokens - 1) // seq_len + eligible = bbc > 0 + self._eligible_shards = np.nonzero(eligible)[0].astype(np.int64) + self._base_block_counts = bbc[self._eligible_shards].astype(np.int64) + def _sample_global_windows(self) -> list[tuple[int, int]]: + assert self._cfg is not None and self._eligible_shards is not None + _, seq_len, _, gns = self._cfg + ec = int(self._eligible_shards.size) + progress = min(self._batches_built / 1800.0, 1.0) + remaining = np.empty(ec, dtype=np.float64) + for i, si in enumerate(self._eligible_shards.tolist()): + if self._cursor_init[si]: + r = int(self._cursor_block_count[si]) - int(self._cursor_next[si]) + remaining[i] = float(max(r, 1)) + else: + remaining[i] = float(self._base_block_counts[i]) + alpha = 0.90 - 0.40 * progress + weights = np.power(remaining, alpha) + ws = float(weights.sum()) + if not np.isfinite(ws) or ws <= 0.0: + weights = np.ones(ec, dtype=np.float64) + ws = float(weights.sum()) + probs = weights / ws + low = min(max(8, self.world_size), ec, gns) + high = min(max(32, self.world_size * 8), ec, gns) + mix = max(1, min(int(round(low + progress * (high - low))), ec, gns)) + cp = self._rng.choice(ec, size=mix, replace=False, p=probs) + cs = self._eligible_shards[cp] + cpr = probs[cp].copy() + cpr /= cpr.sum() + counts = np.ones(mix, dtype=np.int64) + extra = gns - mix + if extra > 0: + counts += self._rng.multinomial(extra, cpr).astype(np.int64) + perm = self._rng.permutation(mix) + cs, counts = cs[perm], counts[perm] + buckets: list[list[tuple[int, int]]] = [] + for si, cnt in zip(cs.tolist(), counts.tolist()): + b: list[tuple[int, int]] = [] + self._take_from_shard(int(si), seq_len, int(cnt), b) + if b: + if len(b) > 1: + bp = self._rng.permutation(len(b)) + b = [b[int(k)] for k in bp.tolist()] + buckets.append(b) + windows: list[tuple[int, int]] = [] + active = [i for i, bk in enumerate(buckets) if bk] + while active: + order = self._rng.permutation(len(active)) + new_active: list[int] = [] + for oi in order.tolist(): + bi = active[oi] + if buckets[bi]: + windows.append(buckets[bi].pop()) + if buckets[bi]: + new_active.append(bi) + active = new_active + return windows + def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: + if self._cfg is None: + self._init_pipeline(global_tokens, seq_len, grad_accum_steps) + _, _, num_seqs, gns = self._cfg + gw = self._sample_global_windows() + local_w = gw[self.rank::self.world_size] + x = torch.empty((num_seqs, seq_len), dtype=torch.int64) + y = torch.empty((num_seqs, seq_len), dtype=torch.int64) + for slot, (si, pos) in enumerate(local_w): + mm = _get_shard_memmap(self.files[si]) + window = torch.as_tensor(np.array(mm[pos:pos + seq_len + 1], dtype=np.int64)) + x[slot] = window[:-1] + y[slot] = window[1:] + self._batches_built += 1 + return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) + +# --- Transformer modules --- + +class RMSNorm(nn.Module): + def __init__(self, eps: float | None = None): + super().__init__() + self.eps = eps + def forward(self, x: Tensor) -> Tensor: + return F.rms_norm(x, (x.size(-1),), eps=self.eps) +class CastedLinear(nn.Linear): + def forward(self, x: Tensor) -> Tensor: + w = self.weight.to(x.dtype) + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) +def restore_low_dim_params_to_fp32(module: nn.Module) -> None: + with torch.no_grad(): + for name, param in module.named_parameters(): + if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: + param.data = param.data.float() +class Rotary(nn.Module): + def __init__(self, dim: int, base: float = 10000.0, train_seq_len: int = 1024, rope_dims: int = 0): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / (base ** (torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached: Tensor | None = None + self._sin_cached: Tensor | None = None + def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached != seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * (scale ** (rd / (rd - 2))) + inv_freq = 1.0 / (new_base ** (torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd)) + else: + inv_freq = self.inv_freq.to(device) + t = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor, rope_dims: int = 0) -> Tensor: + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + +class CausalSelfAttention(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + rope_base: float, + qk_gain_init: float, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + # No CastedLinear -- weights come from banks + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 # set by GPT.__init__ for partial RoPE + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False # set by GPT.__init__ for deep layers only + def _xsa_efficient(self, y: Tensor, v: Tensor) -> Tensor: + """Efficient XSA: subtract self-value projection via GQA-aware reshape (no repeat_interleave). + y: [B, T, H, D], v: [B, T, Hkv, D]. H must be divisible by Hkv.""" + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) # [B, T, Hkv, group, D] + vn = F.normalize(v, dim=-1).unsqueeze(-2) # [B, T, Hkv, 1, D] -- broadcast ready + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, v_embed: Tensor | None = None) -> Tensor: + if getattr(self, '_save_gptq', False): + self._gptq_qkv_in = x.detach() + bsz, seqlen, dim = x.shape + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, v_w.to(x.dtype)) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + y = flash_attn_3_func(q, k, v, causal=True) + if self.use_xsa: + y = self._xsa_efficient(y, v) + y = y.reshape(bsz, seqlen, dim) + if getattr(self, '_save_gptq', False): + self._gptq_o_in = y.detach() + return F.linear(y, out_w.to(x.dtype)) + +class SmearGate(nn.Module): + def __init__(self, dim: int): + super().__init__() + self.gate = nn.Parameter(torch.zeros(dim, dtype=torch.float32)) + def forward(self, x: Tensor) -> Tensor: + g = torch.sigmoid(self.gate.to(dtype=x.dtype))[None, None, :] + x_prev = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) + return (1 - g) * x + g * x_prev + +class BigramHashEmbedding(nn.Module): + def __init__(self, bigram_vocab_size: int, bigram_dim: int, model_dim: int): + super().__init__() + self.bigram_vocab_size = bigram_vocab_size + self.embed = nn.Embedding(bigram_vocab_size, bigram_dim) + nn.init.zeros_(self.embed.weight) + self.proj = CastedLinear(bigram_dim, model_dim, bias=False) if bigram_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.05, dtype=torch.float32)) + def bigram_hash(self, tokens: Tensor) -> Tensor: + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., 0] = mod + out[..., 1:] = torch.bitwise_xor(36313 * t[..., 1:], 27191 * t[..., :-1]) % mod + return out.long() + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(self.bigram_hash(token_ids)) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class ValueEmbedding(nn.Module): + """Reinject token identity into attention values at specific layers. + Each table maps vocab tokens to a low-dim embedding, projected to model_dim.""" + def __init__(self, vocab_size: int, ve_dim: int, model_dim: int): + super().__init__() + self.embed = nn.Embedding(vocab_size, ve_dim) + nn.init.normal_(self.embed.weight, std=0.01) + self.proj = CastedLinear(ve_dim, model_dim, bias=False) if ve_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.1, dtype=torch.float32)) + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(token_ids) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class MLP(nn.Module): + def __init__(self, dim: int, mlp_mult: int, neg_slope: float = 0.5): + super().__init__() + self.neg_slope = neg_slope + # No CastedLinear -- weights come from banks + def forward(self, x: Tensor, up_w: Tensor, down_w: Tensor) -> Tensor: + if getattr(self, '_save_gptq', False): + self._gptq_up_in = x.detach() + x = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=self.neg_slope) + x2 = x.square() + if getattr(self, '_save_gptq', False): + self._gptq_down_in = x2.detach() + return F.linear(x2, down_w.to(x.dtype)) + +class Block(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + rope_base: float, + qk_gain_init: float, + layer_idx: int = 0, + ln_scale: bool = False, + neg_slope: float = 0.5, + ): + super().__init__() + self.layer_idx = layer_idx + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = MLP(dim, mlp_mult, neg_slope=neg_slope) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + def forward(self, x: Tensor, x0: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, up_w: Tensor, down_w: Tensor, v_embed: Tensor | None = None) -> Tensor: + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, q_w, k_w, v_w, out_w, v_embed=v_embed) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + mlp_out = self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + return x_out + mlp_out + +class GPT(nn.Module): + def __init__( + self, + vocab_size: int, + num_layers: int, + model_dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + tie_embeddings: bool, + tied_embed_init_std: float, + logit_softcap: float, + rope_base: float, + qk_gain_init: float, + bigram_vocab_size: int = 0, + bigram_dim: int = 128, + xsa_last_n: int = 0, + rope_dims: int = 0, + ln_scale: bool = False, + ve_enabled: bool = False, + ve_dim: int = 128, + ve_layers: str = "9,10", + neg_slope: float = 0.5, + ): + super().__init__() + self._ve_target_dim = num_kv_heads * (model_dim // num_heads) # kv_dim for value projection + if logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") + self.tie_embeddings = tie_embeddings + self.tied_embed_init_std = tied_embed_init_std + self.logit_softcap = logit_softcap + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + # Parameter banks: contiguous 3D tensors for batched optimizer + head_dim = model_dim // num_heads + kv_dim = num_kv_heads * head_dim + mlp_dim = int(mlp_mult * model_dim) + self.num_layers = num_layers + self.qo_bank = nn.Parameter(torch.empty(2 * num_layers, model_dim, model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * num_layers, kv_dim, model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(num_layers, mlp_dim, model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(num_layers, model_dim, mlp_dim)) + self.blocks = nn.ModuleList( + [ + Block( + model_dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + layer_idx=i, + ln_scale=ln_scale, + neg_slope=neg_slope, + ) + for i in range(num_layers) + ] + ) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + kv_dim_ve = self._ve_target_dim + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim_ve) + self.ve_layer_scales = nn.ParameterList( + [nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices] + ) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.value_embeds = nn.ModuleList() # keep empty for compat + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + if self.lm_head is not None: + self.lm_head._zero_init = True + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + self._init_weights() + def _init_weights(self) -> None: + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + # Init banks: orthogonal, with proj layers scaled down and out/down zero-init + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) # Q + nn.init.zeros_(self.qo_bank.data[n + i]) # Out (zero init) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) # K + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) # V + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) # MLP up + nn.init.zeros_(self.mlp_down_bank.data[i]) # MLP down (zero init) + # Scale proj layers (out_proj and mlp_down are "proj" layers) + self.qo_bank.data[n + i].mul_(proj_scale) + self.mlp_down_bank.data[i].mul_(proj_scale) + # Init remaining nn.Linear modules (bigram proj, lm_head) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif module.weight.ndim == 2 and module.weight.shape[0] >= 64 and module.weight.shape[1] >= 64: + nn.init.orthogonal_(module.weight, gain=1.0) + def _get_ve(self, layer_idx: int, input_ids: Tensor, ve_cache: dict | None = None) -> Tensor | None: + """Get value embedding for a specific layer using shared table + per-layer scale.""" + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if ve_cache is not None and 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_base = ve_cache['ve'] if ve_cache is not None else self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_base * self.ve_layer_scales[ve_idx].to(dtype=ve_base.dtype) + def forward(self, input_ids: Tensor, target_ids: Tensor) -> Tensor: + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + if self.tie_embeddings: + logits_proj = F.linear(x_flat, self.tok_emb.weight) + else: + if self.lm_head is None: + raise RuntimeError("lm_head is required when tie_embeddings=False") + logits_proj = self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy(logits.float(), targets, reduction="mean") + def forward_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve) + x = self.final_norm(x) + if self.tie_embeddings: + logits_proj = F.linear(x, self.tok_emb.weight) + else: + logits_proj = self.lm_head(x) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + +# --- Sliding window evaluation --- + +def eval_val_sliding( + args: Hyperparameters, + base_model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + stride: int, + batch_seqs: int = 32, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + """Sliding window evaluation: each token scored with maximum context.""" + seq_len = eval_seq_len or args.train_seq_len + total_tokens = val_tokens.numel() - 1 + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= 1] + total_windows = len(window_starts) + my_s = (total_windows * rank) // world_size + my_e = (total_windows * (rank + 1)) // world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + base_model.eval() + compiled_logits = torch.compile(base_model.forward_logits, dynamic=False, fullgraph=True) + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk[:-1] + y_batch[i, :wlen] = chunk[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = compiled_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), + reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt = y_batch[i, s:wlen] + prev = x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + val_loss = (loss_sum / token_count).item() + bits_per_token = val_loss / math.log(2.0) + tokens_per_byte = token_count.item() / byte_count.item() + base_model.train() + return val_loss, bits_per_token * tokens_per_byte + + +def eval_val_sliding_ttt( + args: Hyperparameters, base_model: nn.Module, rank: int, world_size: int, + device: torch.device, val_tokens: Tensor, base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, is_boundary_token_lut: Tensor, + stride: int, batch_seqs: int = 32, log0=print, +) -> tuple[float, float]: + """Legal score-first TTT (PR #461 recipe): score each chunk with sliding windows, + then train on it. Every token scored BEFORE any update that could use it.""" + seq_len = args.train_seq_len + total_tokens = val_tokens.numel() - 1 + ttt_chunk = args.ttt_chunk_tokens + + # Pre-compute all window starts + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= stride or ws == 0] + + # Assign each window to a chunk based on the first token it scores + num_chunks = (total_tokens + ttt_chunk - 1) // ttt_chunk + chunk_windows: list[list[int]] = [[] for _ in range(num_chunks)] + for ws in window_starts: + end = min(ws + seq_len, total_tokens) + wlen = end - ws + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_start = ws + s + ci = min(scored_start // ttt_chunk, num_chunks - 1) + chunk_windows[ci].append(ws) + + log0(f"ttt_sliding:start chunks={num_chunks} chunk_tokens={ttt_chunk} " + f"total_windows={len(window_starts)} stride={stride} " + f"ttt_lr={args.ttt_lr} ttt_epochs={args.ttt_epochs} " + f"freeze_blocks={args.ttt_freeze_blocks}") + + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + + # Freeze first N blocks + frozen_block_ids = set(range(min(args.ttt_freeze_blocks, len(base_model.blocks)))) + ttt_params = [] + for name, p in base_model.named_parameters(): + freeze = False + for bi in frozen_block_ids: + if f"blocks.{bi}." in name: + freeze = True + break + if freeze: + p.requires_grad_(False) + else: + p.requires_grad_(True) + ttt_params.append(p) + + log0(f"ttt_sliding:params unfrozen={sum(p.numel() for p in ttt_params)} " + f"frozen={sum(p.numel() for p in base_model.parameters() if not p.requires_grad)}") + + optimizer = torch.optim.SGD(ttt_params, lr=args.ttt_lr, momentum=args.ttt_momentum) + t0 = time.perf_counter() + + for ci in range(num_chunks): + windows = chunk_windows[ci] + if not windows: + continue + chunk_start = ci * ttt_chunk + chunk_end = min((ci + 1) * ttt_chunk, total_tokens) + + # --- Phase 1: SCORE this chunk's windows (inference_mode) --- + my_s = (len(windows) * rank) // world_size + my_e = (len(windows) * (rank + 1)) // world_size + my_windows = windows[my_s:my_e] + + base_model.eval() + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk_tok = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk_tok[:-1] + y_batch[i, :wlen] = chunk_tok[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = base_model.forward_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt, prev = y_batch[i, s:wlen], x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + + # --- Phase 2: TRAIN on this chunk (already scored = legal) --- + is_last_chunk = (ci == num_chunks - 1) + if not is_last_chunk and args.ttt_epochs > 0: + base_model.train() + chunk_seqs = (chunk_end - chunk_start) // seq_len + if chunk_seqs > 0: + cos_lr = args.ttt_lr * 0.5 * (1.0 + math.cos(math.pi * ci / max(num_chunks - 1, 1))) + for pg in optimizer.param_groups: + pg['lr'] = cos_lr + my_seq_s = (chunk_seqs * rank) // world_size + my_seq_e = (chunk_seqs * (rank + 1)) // world_size + my_chunk_seqs = my_seq_e - my_seq_s + for _ep in range(args.ttt_epochs): + for bs in range(0, my_chunk_seqs, args.ttt_batch_seqs): + be = min(bs + args.ttt_batch_seqs, my_chunk_seqs) + actual_bs = my_seq_s + bs + start_tok = chunk_start + actual_bs * seq_len + end_tok = chunk_start + (my_seq_s + be) * seq_len + 1 + if end_tok > val_tokens.numel(): + continue + local = val_tokens[start_tok:end_tok].to(device=device, dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + optimizer.zero_grad(set_to_none=True) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + loss = base_model(x, y) + loss.backward() + if world_size > 1: + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + torch.nn.utils.clip_grad_norm_(ttt_params, args.ttt_grad_clip) + optimizer.step() + + if rank == 0 and (ci % 10 == 0 or ci == num_chunks - 1): + elapsed = time.perf_counter() - t0 + rl = loss_sum.item() / max(token_count.item(), 1) + rbpb = rl / math.log(2.0) * (token_count.item() / max(byte_count.item(), 1)) if token_count.item() > 0 else 0.0 + log0(f" ttt_chunk [{ci+1}/{num_chunks}] bpb={rbpb:.6f} time={elapsed:.1f}s") + + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_count.item()) + + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + + log0(f"ttt_sliding:done val_loss={val_loss:.6f} val_bpb={val_bpb:.6f} " + f"elapsed={time.perf_counter() - t0:.1f}s") + return val_loss, val_bpb + + +# --- GPTQ-lite int6 quantization --- + +def _classify_param(name: str) -> str: + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or (".proj." in name and ".mlp." not in name): + return "attn" + return "other" +def quantize_int6_per_row(t: Tensor, clip_range: int = 31) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + best_q, best_s, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(t32.abs(), pct, dim=1) + else: + row_clip = t32.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + recon = q.float() * s.float()[:, None] + err = (t32 - recon).pow(2).mean().item() + if err < best_err: + best_q, best_s, best_err = q, s, err + return best_q, best_s + amax = t32.abs().max().item() + scale = torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + q = torch.clamp(torch.round(t32 / scale.float()), -clip_range, clip_range).to(torch.int8) + return q, scale + +def _unbank_state_dict(sd: dict[str, Tensor], num_layers: int) -> dict[str, Tensor]: + """Convert 3D bank tensors into individual 2D tensors with standard names.""" + out: dict[str, Tensor] = {} + n = num_layers + for name, tensor in sd.items(): + if name == "qo_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_q.weight"] = tensor[i] + out[f"blocks.{i}.attn.proj.weight"] = tensor[n + i] + elif name == "kv_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_k.weight"] = tensor[i] + out[f"blocks.{i}.attn.c_v.weight"] = tensor[n + i] + elif name == "mlp_up_bank": + for i in range(n): + out[f"blocks.{i}.mlp.fc.weight"] = tensor[i] + elif name == "mlp_down_bank": + for i in range(n): + out[f"blocks.{i}.mlp.proj.weight"] = tensor[i] + else: + out[name] = tensor + return out + +def _rebank_state_dict(sd: dict[str, Tensor], num_layers: int, template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + """Convert individual 2D tensors back into 3D bank tensors.""" + out: dict[str, Tensor] = {} + n = num_layers + # Reconstruct banks from individual weight keys + qo_slices = [None] * (2 * n) + kv_slices = [None] * (2 * n) + up_slices = [None] * n + down_slices = [None] * n + consumed = set() + for i in range(n): + qk = f"blocks.{i}.attn.c_q.weight" + if qk in sd: + qo_slices[i] = sd[qk] + consumed.add(qk) + ok = f"blocks.{i}.attn.proj.weight" + if ok in sd: + qo_slices[n + i] = sd[ok] + consumed.add(ok) + kk = f"blocks.{i}.attn.c_k.weight" + if kk in sd: + kv_slices[i] = sd[kk] + consumed.add(kk) + vk = f"blocks.{i}.attn.c_v.weight" + if vk in sd: + kv_slices[n + i] = sd[vk] + consumed.add(vk) + fk = f"blocks.{i}.mlp.fc.weight" + if fk in sd: + up_slices[i] = sd[fk] + consumed.add(fk) + dk = f"blocks.{i}.mlp.proj.weight" + if dk in sd: + down_slices[i] = sd[dk] + consumed.add(dk) + out["qo_bank"] = torch.stack(qo_slices).to(dtype=template_sd["qo_bank"].dtype) + out["kv_bank"] = torch.stack(kv_slices).to(dtype=template_sd["kv_bank"].dtype) + out["mlp_up_bank"] = torch.stack(up_slices).to(dtype=template_sd["mlp_up_bank"].dtype) + out["mlp_down_bank"] = torch.stack(down_slices).to(dtype=template_sd["mlp_down_bank"].dtype) + for name, tensor in sd.items(): + if name not in consumed: + out[name] = tensor + return out + +def mixed_quantize_int6(state_dict: dict[str, Tensor], int6_cats: set[str], clip_range: int = 31, + hessians: dict[str, Tensor] | None = None): + num_layers_total = max( + (int(k.split(".")[1]) for k in state_dict if k.startswith("blocks.")), + default=0, + ) + 1 + late_k_layers = set(range(num_layers_total - 2, num_layers_total)) + result: dict[str, Tensor] = {} + meta: dict[str, object] = {} + gptq_count, naive_count = 0, 0 + for name, tensor in state_dict.items(): + t = tensor.detach().cpu().contiguous() + cat = _classify_param(name) + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough" + continue + if any(p in name for p in CONTROL_TENSOR_NAME_PATTERNS): + result[name] = t.float() + meta[name] = "passthrough_ctrl" + continue + if cat in int6_cats and t.ndim >= 1: + H = hessians.get(name) if hessians else None + if H is not None and t.ndim == 2: + q, s = gptq_quantize_weight(t, H.cpu(), clip_range=clip_range) + gptq_count += 1 + else: + q, s = quantize_int6_per_row(t, clip_range=clip_range) + naive_count += 1 + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int6"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + if hessians: + print(f"gptq_quantize: {gptq_count} GPTQ layers, {naive_count} naive layers", flush=True) + return result, meta +def dequantize_mixed_int6(result: dict[str, Tensor], meta: dict[str, object], + template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + for name, orig in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if info in ("passthrough", "passthrough_ctrl", "passthrough_fp16"): + t = result[name] + if t.dtype == torch.float16 and orig_dtype in (torch.float32, torch.bfloat16): + t = t.to(orig_dtype) + out[name] = t + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out + +# --- Full Hessian GPTQ --- + +def gptq_quantize_weight(W: Tensor, H: Tensor, clip_range: int = 31, + block_size: int = 128, percdamp: float = 0.01) -> tuple[Tensor, Tensor]: + """GPTQ with Cholesky error compensation and actorder (Frantar et al., ICLR 2023).""" + W_orig = W.float().clone() + rows, cols = W_orig.shape + H = H.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = percdamp * H.diag().mean() + H.diagonal().add_(damp) + perm = torch.argsort(H.diag(), descending=True) + invperm = torch.argsort(perm) + W_perm = W_orig[:, perm].clone() + W_perm[:, dead[perm]] = 0 + H = H[perm][:, perm] + try: + Hinv = torch.cholesky_inverse(torch.linalg.cholesky(H)) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + except torch.linalg.LinAlgError: + return quantize_int6_per_row(W_orig, clip_range) + best_q, best_scale, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(W_orig.abs(), pct, dim=1) + else: + row_clip = W_orig.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + sf = s.float() + Q = torch.zeros(rows, cols, dtype=torch.int8) + W_work = W_perm.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + W_block = W_work[:, i1:i2].clone() + Hinv_block = Hinv[i1:i2, i1:i2] + Err = torch.zeros(rows, i2 - i1) + for j in range(i2 - i1): + w_col = W_block[:, j] + d = Hinv_block[j, j] + q_col = torch.clamp(torch.round(w_col / sf), -clip_range, clip_range) + Q[:, i1 + j] = q_col.to(torch.int8) + err = (w_col - q_col.float() * sf) / d + Err[:, j] = err + W_block[:, j:] -= err.unsqueeze(1) * Hinv_block[j, j:].unsqueeze(0) + if i2 < cols: + W_work[:, i2:] -= Err @ Hinv[i1:i2, i2:] + recon = Q.float() * sf[:, None] + mse = (W_perm - recon).pow(2).mean().item() + if mse < best_err: + best_q, best_scale, best_err = Q, s, mse + best_q = best_q[:, invperm] + return best_q, best_scale + +def _init_hessians(nl: int, dim: int, mlp_dim: int, device: torch.device) -> dict[str, Tensor]: + h: dict[str, Tensor] = {} + for i in range(nl): + for k in ['c_q', 'c_k', 'c_v']: + h[f'blocks.{i}.attn.{k}.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.attn.proj.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.mlp.fc.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.mlp.proj.weight'] = torch.zeros(mlp_dim, mlp_dim, dtype=torch.float32, device=device) + return h + +def _accum_hessians(hessians: dict[str, Tensor], blocks: nn.ModuleList, dim: int, mlp_dim: int) -> None: + for i, block in enumerate(blocks): + qkv_in = block.attn._gptq_qkv_in.float().reshape(-1, dim) + h_qkv = qkv_in.t() @ qkv_in + hessians[f'blocks.{i}.attn.c_q.weight'] += h_qkv + hessians[f'blocks.{i}.attn.c_k.weight'] += h_qkv + hessians[f'blocks.{i}.attn.c_v.weight'] += h_qkv + o_in = block.attn._gptq_o_in.float().reshape(-1, dim) + hessians[f'blocks.{i}.attn.proj.weight'] += o_in.t() @ o_in + up_in = block.mlp._gptq_up_in.float().reshape(-1, dim) + hessians[f'blocks.{i}.mlp.fc.weight'] += up_in.t() @ up_in + down_in = block.mlp._gptq_down_in.float().reshape(-1, mlp_dim) + hessians[f'blocks.{i}.mlp.proj.weight'] += down_in.t() @ down_in + +def _finalize_hessians(hessians: dict[str, Tensor], num_batches: int) -> None: + for name in hessians: + hessians[name] = hessians[name].cpu() / num_batches + damp = 0.01 * torch.diag(hessians[name]).mean().clamp_min(1e-6) + hessians[name] += damp * torch.eye(hessians[name].shape[0]) + +def gptq_collect_hessians(base_model: nn.Module, train_loader, device: torch.device, + num_batches: int, batch_tokens: int, seq_len: int, + grad_accum_steps: int) -> dict[str, Tensor]: + """Collect Hessians H = X^T X from training data.""" + nl = base_model.num_layers + dim = base_model.tok_emb.weight.shape[1] + mlp_dim = base_model.mlp_up_bank.shape[1] + hessians = _init_hessians(nl, dim, mlp_dim, device) + for block in base_model.blocks: + block.attn._save_gptq = True + block.mlp._save_gptq = True + base_model.eval() + with torch.inference_mode(), torch.autocast(device_type='cuda', dtype=torch.bfloat16): + for _ in range(num_batches): + x, y = train_loader.next_batch(batch_tokens, seq_len, grad_accum_steps) + base_model(x, y) + _accum_hessians(hessians, base_model.blocks, dim, mlp_dim) + for block in base_model.blocks: + block.attn._save_gptq = False + block.mlp._save_gptq = False + _finalize_hessians(hessians, num_batches) + base_model.train() + return hessians + +# --- Training --- + +def main() -> None: + code = Path(__file__).read_text(encoding="utf-8") + args = Hyperparameters() + # zeropower_via_newtonschulz5 runs eagerly with bmm -- do NOT compile + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError(f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral") + grad_accum_steps = 8 // world_size + grad_scale = 1.0 / grad_accum_steps + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + master_process = rank == 0 + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + logfile = None + if master_process: + os.makedirs("logs", exist_ok=True) + logfile = f"logs/{args.run_id}.txt" + print(logfile) + def log0(msg: str, console: bool = True) -> None: + if not master_process: + return + if console: + print(msg) + if logfile is not None: + with open(logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + log0(code, console=False) + log0("=" * 100, console=False) + log0(f"Running Python {sys.version}", console=False) + log0(f"Running PyTorch {torch.__version__}", console=False) + log0( + subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=False).stdout, + console=False, + ) + log0("=" * 100, console=False) + random.seed(args.seed) + np.random.seed(args.seed) + torch.manual_seed(args.seed) + torch.cuda.manual_seed_all(args.seed) + (base_bytes_lut, has_leading_space_lut, is_boundary_token_lut), tokenizer_metadata = load_tokenizer_luts( + args.tokenizer_path, args.tokenizer_meta_path, args.vocab_size, device, + validate_meta=args.tokenizer_meta_validate, + ) + log0(f"tokenizer: kind={tokenizer_metadata.get('tokenizer_kind', 'unknown')} vocab={tokenizer_metadata.get('vocab_size', '?')}") + if tokenizer_metadata.get('tokenizer_kind') == 'sentencepiece': + sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) + if int(sp.vocab_size()) != args.vocab_size: + raise ValueError( + f"VOCAB_SIZE={args.vocab_size} does not match tokenizer vocab_size={int(sp.vocab_size())}" + ) + dataset_dir = Path(args.data_path).resolve() + actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) + effective_eval_seq_len = args.eval_seq_len if args.eval_seq_len > 0 else args.train_seq_len + val_seq_len = max(args.train_seq_len, effective_eval_seq_len) + val_tokens = load_validation_tokens(args.val_files, val_seq_len) + log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") + log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") + base_model = GPT( + vocab_size=args.vocab_size, + num_layers=args.num_layers, + model_dim=args.model_dim, + num_heads=args.num_heads, + num_kv_heads=args.num_kv_heads, + mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, + tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, + rope_base=args.rope_base, + qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, + bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, + ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, + ve_dim=args.ve_dim, + ve_layers=args.ve_layers, + neg_slope=args.negative_slope, + ).to(device).bfloat16() + # Banks stay FP32 (like CastedLinear weights), cast to BF16 in forward + base_model.qo_bank.data = base_model.qo_bank.data.float() + base_model.kv_bank.data = base_model.kv_bank.data.float() + base_model.mlp_up_bank.data = base_model.mlp_up_bank.data.float() + base_model.mlp_down_bank.data = base_model.mlp_down_bank.data.float() + for module in base_model.modules(): + if isinstance(module, CastedLinear): + module.float() + restore_low_dim_params_to_fp32(base_model) + # No DDP -- Parallel Muon handles bank grad communication via reduce-scatter, + # and non-bank grads are manually all-reduced before Adam steps. + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + model = compiled_model + + # Optimizer split: + # - 4 parameter banks -> Muon (batched Newton-Schulz) + # - token embedding -> Adam + # - scalars/control tensors -> Adam + # - bigram proj, VE proj -> Adam (small matrix params not worth banking) + matrix_params = [ + base_model.qo_bank, base_model.kv_bank, + base_model.mlp_up_bank, base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for name, p in block_named_params + if p.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + scalar_params.append(base_model.smear.gate) + if base_model.bigram is not None: + scalar_params.append(base_model.bigram.scale) + token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr + tok_params = [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}] + if base_model.bigram is not None: + tok_params.append({"params": [base_model.bigram.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.bigram.proj is not None: + scalar_params.append(base_model.bigram.proj.weight) + if base_model.ve_shared is not None: + tok_params.append({"params": [base_model.ve_shared.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.ve_shared.proj is not None: + scalar_params.append(base_model.ve_shared.proj.weight) + scalar_params.append(base_model.ve_shared.scale) + for s in base_model.ve_layer_scales: + scalar_params.append(s) + optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + optimizer_muon = Muon( + matrix_params, + lr=args.matrix_lr, + momentum=args.muon_momentum, + backend_steps=args.muon_backend_steps, + weight_decay=args.muon_wd, + ) + for group in optimizer_muon.param_groups: + group["base_lr"] = args.matrix_lr + optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + # Non-bank params that need manual all-reduce (replicated across GPUs) + replicated_params = list(optimizer_tok.param_groups[0]["params"]) + for pg in optimizer_tok.param_groups[1:]: + replicated_params.extend(pg["params"]) + replicated_params.extend(scalar_params) + + optimizer_head = None + if base_model.lm_head is not None: + optimizer_head = torch.optim.Adam( + [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + replicated_params.append(base_model.lm_head.weight) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if optimizer_head is not None: + optimizers.append(optimizer_head) + log0(f"model_params:{sum(p.numel() for p in base_model.parameters())}") + xsa_layers = [i for i, b in enumerate(base_model.blocks) if b.attn.use_xsa] + log0(f"XSA:last_{args.xsa_last_n} active_layers:{xsa_layers}") + log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") + log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") + log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") + log0( + f"tie_embeddings:{args.tie_embeddings} embed_lr:{token_lr} " + f"head_lr:{args.head_lr if base_model.lm_head is not None else 0.0} " + f"matrix_lr:{args.matrix_lr} scalar_lr:{args.scalar_lr}" + ) + log0( + f"train_batch_tokens:{args.train_batch_tokens} train_seq_len:{args.train_seq_len} " + f"iterations:{args.iterations} warmup_steps:{args.warmup_steps} " + f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" + ) + log0(f"seed:{args.seed}") + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + def zero_grad_all() -> None: + for opt in optimizers: + opt.zero_grad(set_to_none=True) + max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None + if args.use_gptq and max_wallclock_ms is not None: + max_wallclock_ms -= args.gptq_reserve_ms + log0(f"gptq:reserving {args.gptq_reserve_ms:.0f}ms from training budget, effective={max_wallclock_ms:.0f}ms") + def lr_mul(step: int, elapsed_ms: float) -> float: + if args.warmdown_iters <= 0: + return 1.0 + if max_wallclock_ms is None: + warmdown_start = max(args.iterations - args.warmdown_iters, 0) + return max((args.iterations - step) / max(args.warmdown_iters, 1), 0.0) if warmdown_start <= step < args.iterations else 1.0 + step_ms = elapsed_ms / max(step, 1) + warmdown_ms = args.warmdown_iters * step_ms + remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) + return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 + if args.warmup_steps > 0: + initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} + initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] + model.train() + for warmup_step in range(args.warmup_steps): + zero_grad_all() + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + warmup_loss = model(x, y) + (warmup_loss * grad_scale).backward() + # All-reduce all grads for warmup (simple, not optimized) + if distributed: + for p in base_model.parameters(): + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + for opt in optimizers: + opt.step() + zero_grad_all() + if args.warmup_steps <= 20 or (warmup_step + 1) % 10 == 0 or warmup_step + 1 == args.warmup_steps: + log0(f"warmup_step:{warmup_step + 1}/{args.warmup_steps}") + base_model.load_state_dict(initial_model_state, strict=True) + for opt, state in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + zero_grad_all() + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + swa_state: dict[str, Tensor] | None = None + swa_count = 0 + ema_state = {name: t.detach().float().clone() for name, t in base_model.state_dict().items()} + ema_decay = 0.997 + training_time_ms = 0.0 + stop_after_step: int | None = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) + should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1000.0 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + log0( + f"step:{step}/{args.iterations} val_loss:{val_loss:.4f} val_bpb:{val_bpb:.4f} " + f"train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms / max(step, 1):.2f}ms" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < args.iterations: + log0( + f"stopping_early: wallclock_cap train_time:{training_time_ms:.0f}ms " + f"step:{step}/{args.iterations}" + ) + break + elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + scale = lr_mul(step, elapsed_ms) + zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y) + train_loss += loss.detach() + (loss * grad_scale).backward() + train_loss /= grad_accum_steps + frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 + muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum + for group in optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * scale + if args.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) + # === 3-phase overlapped optimizer step === + # Phase 1: Launch async reduce-scatter for banks (biggest first) + optimizer_muon.launch_reduce_scatters() + # Phase 2: All-reduce non-bank grads + step Adam (while bank RS is in-flight) + if distributed: + for p in replicated_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + optimizer_tok.step() + optimizer_scalar.step() + if optimizer_head is not None: + optimizer_head.step() + # Phase 3: Wait for RS, local NS5, all-gather (banks processed last) + optimizer_muon.step() + zero_grad_all() + # EMA update + with torch.no_grad(): + for name, t in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_(t.detach().float(), alpha=1.0 - ema_decay) + step += 1 + approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + if args.swa_enabled and scale < 0.2 and step % args.swa_every == 0: + if swa_state is None: + swa_state = {name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()} + swa_count = 1 + log0(f"swa:start step:{step}") + else: + for name, t in base_model.state_dict().items(): + swa_state[name] += t.detach().cpu() + swa_count += 1 + should_log_train = ( + args.train_log_every > 0 + and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) + ) + if should_log_train: + log0( + f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " + f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" + ) + reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + if distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log0( + f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " + f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" + ) + # Apply EMA weights + log0("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = {name: t.to(dtype=current_state[name].dtype) for name, t in ema_state.items()} + base_model.load_state_dict(avg_state, strict=True) + torch.cuda.synchronize() + t_diag = time.perf_counter() + diag_val_loss, diag_val_bpb = eval_val( + args, compiled_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"DIAGNOSTIC post_ema val_loss:{diag_val_loss:.4f} val_bpb:{diag_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_diag):.0f}ms" + ) + export_sd = base_model.state_dict() + if master_process: + torch.save(export_sd, "final_model.pt") + model_bytes = os.path.getsize("final_model.pt") + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model: {model_bytes} bytes") + log0(f"Code size: {code_bytes} bytes") + # Unbank 3D tensors into individual 2D tensors for quantization + sd_cpu = {k: v.detach().cpu() for k, v in export_sd.items()} + unbanked_sd = _unbank_state_dict(sd_cpu, args.num_layers) + # GPTQ calibration: collect Hessians from training data + gptq_hessians = None + if args.use_gptq: + t_gptq = time.perf_counter() + log0(f"gptq:calibrating with {args.gptq_calib_samples} batches (training data)...") + calib_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + gptq_hessians = gptq_collect_hessians( + base_model, calib_loader, device, num_batches=args.gptq_calib_samples, + batch_tokens=args.train_batch_tokens, seq_len=args.train_seq_len, + grad_accum_steps=grad_accum_steps) + del calib_loader + gptq_elapsed = time.perf_counter() - t_gptq + log0(f"gptq:calibrated {len(gptq_hessians)} layers in {gptq_elapsed:.1f}s") + torch.cuda.empty_cache() + quant_result, quant_meta = mixed_quantize_int6(unbanked_sd, {"mlp", "attn"}, clip_range=args.quant_clip_range, hessians=gptq_hessians) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = lzma.compress(quant_raw, preset=6) + if master_process: + with open("final_model.int6.ptz", "wb") as f: + f.write(quant_blob) + quant_file_bytes = len(quant_blob) + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model int6+lzma: {quant_file_bytes} bytes") + log0(f"Total submission size int6+lzma: {quant_file_bytes + code_bytes} bytes") + if distributed: + dist.barrier() + with open("final_model.int6.ptz", "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(lzma.decompress(quant_blob_disk)), + map_location="cpu", + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd) + # Re-bank the dequantized tensors + deq_state = _rebank_state_dict(deq_unbanked, args.num_layers, sd_cpu) + eval_model = GPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + neg_slope=args.negative_slope, + ).to(device).bfloat16() + eval_model.qo_bank.data = eval_model.qo_bank.data.float() + eval_model.kv_bank.data = eval_model.kv_bank.data.float() + eval_model.mlp_up_bank.data = eval_model.mlp_up_bank.data.float() + eval_model.mlp_down_bank.data = eval_model.mlp_down_bank.data.float() + for m in eval_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(eval_model) + eval_model.load_state_dict(deq_state, strict=True) + compiled_eval = torch.compile(eval_model, dynamic=False, fullgraph=True) + torch.cuda.synchronize() + t_qeval = time.perf_counter() + q_val_loss, q_val_bpb = eval_val( + args, compiled_eval, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + eval_seq_len=effective_eval_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" + ) + log0(f"final_int6_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + sw_seq_len = effective_eval_seq_len + if args.eval_stride > 0 and args.eval_stride < sw_seq_len: + torch.cuda.synchronize() + t_slide = time.perf_counter() + sw_val_loss, sw_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window val_loss:{sw_val_loss:.4f} val_bpb:{sw_val_bpb:.4f} " + f"stride:{args.eval_stride} eval_time:{1000.0 * (time.perf_counter() - t_slide):.0f}ms" + ) + log0(f"final_int6_sliding_window_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + if args.eval_stride != 64 and 64 < sw_seq_len: + torch.cuda.synchronize() + t_slide64 = time.perf_counter() + sw64_val_loss, sw64_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=64, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window_s64 val_loss:{sw64_val_loss:.4f} val_bpb:{sw64_val_bpb:.4f} " + f"stride:64 eval_time:{1000.0 * (time.perf_counter() - t_slide64):.0f}ms" + ) + log0(f"final_int6_sliding_window_s64_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + # Legal score-first TTT (PR #461 recipe) + if args.ttt_enabled: + torch.cuda.synchronize() + t_ttt = time.perf_counter() + ttt_loss, ttt_bpb = eval_val_sliding_ttt( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, log0=log0, + ) + torch.cuda.synchronize() + log0(f"legal_ttt val_loss:{ttt_loss:.4f} val_bpb:{ttt_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt):.0f}ms") + log0(f"legal_ttt_exact val_loss:{ttt_loss:.8f} val_bpb:{ttt_bpb:.8f}") + if distributed: + dist.destroy_process_group() +if __name__ == "__main__": + main() + +==================================================================================================== +Running Python 3.12.3 (main, Mar 3 2026, 12:15:18) [GCC 13.3.0] +Running PyTorch 2.9.1+cu128 +Tue Mar 31 14:43:28 2026 ++-----------------------------------------------------------------------------------------+ +| NVIDIA-SMI 580.126.09 Driver Version: 580.126.09 CUDA Version: 13.0 | ++-----------------------------------------+------------------------+----------------------+ +| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | +| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | +| | | MIG M. | +|=========================================+========================+======================| +| 0 NVIDIA H100 80GB HBM3 On | 00000000:0A:00.0 Off | 0 | +| N/A 32C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 1 NVIDIA H100 80GB HBM3 On | 00000000:18:00.0 Off | 0 | +| N/A 29C P0 124W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 2 NVIDIA H100 80GB HBM3 On | 00000000:3F:00.0 Off | 0 | +| N/A 28C P0 116W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 3 NVIDIA H100 80GB HBM3 On | 00000000:48:00.0 Off | 0 | +| N/A 33C P0 122W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 4 NVIDIA H100 80GB HBM3 On | 00000000:87:00.0 Off | 0 | +| N/A 32C P0 118W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 5 NVIDIA H100 80GB HBM3 On | 00000000:90:00.0 Off | 0 | +| N/A 28C P0 112W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 6 NVIDIA H100 80GB HBM3 On | 00000000:BE:00.0 Off | 0 | +| N/A 28C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 7 NVIDIA H100 80GB HBM3 On | 00000000:C7:00.0 Off | 0 | +| N/A 33C P0 120W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ + ++-----------------------------------------------------------------------------------------+ +| Processes: | +| GPU GI CI PID Type Process name GPU Memory | +| ID ID Usage | +|=========================================================================================| +| 0 N/A N/A 1930 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 1 N/A N/A 1931 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 2 N/A N/A 1932 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 3 N/A N/A 1933 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 4 N/A N/A 1934 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 5 N/A N/A 1935 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 6 N/A N/A 1936 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 7 N/A N/A 1937 C ...ameter-golf/.venv/bin/python3 1512MiB | ++-----------------------------------------------------------------------------------------+ + +==================================================================================================== +tokenizer: kind=tokenmonster vocab=998 +train_loader:dataset:fineweb10B_scylla train_shards:194 +val_loader:shards pattern=./data/datasets/fineweb10B_scylla/fineweb_val_*.bin tokens:62363648 +model_params:27022172 +XSA:last_11 active_layers:[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] +world_size:8 grad_accum_steps:1 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_mode:gqa num_heads:8 num_kv_heads:4 +tie_embeddings:True embed_lr:0.035 head_lr:0.0 matrix_lr:0.025 scalar_lr:0.025 +train_batch_tokens:786432 train_seq_len:2048 iterations:9000 warmup_steps:20 max_wallclock_seconds:600.000 +seed:1337 +gptq:reserving 9000ms from training budget, effective=591000ms +warmup_step:1/20 +warmup_step:2/20 +warmup_step:3/20 +warmup_step:4/20 +warmup_step:5/20 +warmup_step:6/20 +warmup_step:7/20 +warmup_step:8/20 +warmup_step:9/20 +warmup_step:10/20 +warmup_step:11/20 +warmup_step:12/20 +warmup_step:13/20 +warmup_step:14/20 +warmup_step:15/20 +warmup_step:16/20 +warmup_step:17/20 +warmup_step:18/20 +warmup_step:19/20 +warmup_step:20/20 +step:0/9000 val_loss:6.8893 val_bpb:3.3906 train_time:0ms step_avg:0.01ms +step:1/9000 train_loss:6.8885 train_time:133ms step_avg:133.24ms +step:2/9000 train_loss:9.2540 train_time:168ms step_avg:84.25ms +step:3/9000 train_loss:9.1371 train_time:253ms step_avg:84.35ms +step:4/9000 train_loss:8.5749 train_time:338ms step_avg:84.61ms +step:5/9000 train_loss:8.1062 train_time:424ms step_avg:84.76ms +step:6/9000 train_loss:7.5450 train_time:509ms step_avg:84.79ms +step:7/9000 train_loss:7.0905 train_time:594ms step_avg:84.81ms +step:8/9000 train_loss:6.7904 train_time:679ms step_avg:84.83ms +step:9/9000 train_loss:6.6502 train_time:764ms step_avg:84.94ms +step:10/9000 train_loss:6.4416 train_time:849ms step_avg:84.94ms +step:500/9000 train_loss:2.3187 train_time:43818ms step_avg:87.64ms +step:1000/9000 train_loss:2.2047 train_time:87672ms step_avg:87.67ms +step:1500/9000 train_loss:2.1824 train_time:131516ms step_avg:87.68ms +step:2000/9000 train_loss:2.1525 train_time:175431ms step_avg:87.72ms +step:2500/9000 train_loss:2.2113 train_time:219385ms step_avg:87.75ms +step:3000/9000 train_loss:2.1066 train_time:263337ms step_avg:87.78ms +step:3500/9000 train_loss:2.0862 train_time:307270ms step_avg:87.79ms +step:4000/9000 train_loss:2.0368 train_time:351208ms step_avg:87.80ms +step:4000/9000 val_loss:2.0789 val_bpb:1.0232 train_time:351264ms step_avg:87.82ms +step:4500/9000 train_loss:2.0131 train_time:395168ms step_avg:87.82ms +step:5000/9000 train_loss:1.9703 train_time:439082ms step_avg:87.82ms +step:5500/9000 train_loss:1.9399 train_time:483044ms step_avg:87.83ms +step:6000/9000 train_loss:1.9922 train_time:527029ms step_avg:87.84ms +swa:start step:6050 +step:6500/9000 train_loss:1.9783 train_time:571716ms step_avg:87.96ms +step:6716/9000 val_loss:1.9638 val_bpb:0.9665 train_time:591133ms step_avg:88.02ms +stopping_early: wallclock_cap train_time:591133ms step:6716/9000 +peak memory allocated: 23035 MiB reserved: 24046 MiB +ema:applying EMA weights +DIAGNOSTIC post_ema val_loss:1.9622 val_bpb:0.9657 eval_time:2112ms +Serialized model: 106201974 bytes +Code size: 101929 bytes +gptq:calibrating with 64 batches (training data)... +gptq:calibrated 66 layers in 6.7s +Serialized model int6+lzma: 15859076 bytes +Total submission size int6+lzma: 15961005 bytes +final_int6_roundtrip val_loss:1.9683 val_bpb:0.9687 eval_time:16727ms +final_int6_roundtrip_exact val_loss:1.96829674 val_bpb:0.96871918 +final_int6_sliding_window val_loss:1.9285 val_bpb:0.9491 stride:64 eval_time:92334ms +final_int6_sliding_window_exact val_loss:1.92847248 val_bpb:0.94911064 +final_int8_zlib_roundtrip_exact val_loss:1.92847248 val_bpb:0.94911064 +ttt_sliding:start chunks=1904 chunk_tokens=32768 total_windows=974432 stride=64 ttt_lr=0.002 ttt_epochs=3 freeze_blocks=0 +ttt_sliding:params unfrozen=27022172 frozen=0 + ttt_chunk [1/1904] bpb=0.914901 time=0.5s + ttt_chunk [11/1904] bpb=0.913003 time=2.8s + ttt_chunk [21/1904] bpb=0.937304 time=5.0s + ttt_chunk [31/1904] bpb=0.941354 time=7.3s + ttt_chunk [41/1904] bpb=0.968061 time=9.5s + ttt_chunk [51/1904] bpb=0.967601 time=11.8s + ttt_chunk [61/1904] bpb=0.968924 time=14.1s + ttt_chunk [71/1904] bpb=0.962667 time=16.3s + ttt_chunk [81/1904] bpb=0.974603 time=18.6s + ttt_chunk [91/1904] bpb=0.967818 time=20.9s + ttt_chunk [101/1904] bpb=0.965223 time=23.3s + ttt_chunk [111/1904] bpb=0.960300 time=25.6s + ttt_chunk [121/1904] bpb=0.961390 time=28.0s + ttt_chunk [131/1904] bpb=0.961213 time=30.4s + ttt_chunk [141/1904] bpb=0.957305 time=32.7s + ttt_chunk [151/1904] bpb=0.954450 time=35.1s + ttt_chunk [161/1904] bpb=0.953940 time=37.3s + ttt_chunk [171/1904] bpb=0.951811 time=39.6s + ttt_chunk [181/1904] bpb=0.950999 time=41.9s + ttt_chunk [191/1904] bpb=0.950985 time=44.3s + ttt_chunk [201/1904] bpb=0.952431 time=46.7s + ttt_chunk [211/1904] bpb=0.950764 time=49.0s + ttt_chunk [221/1904] bpb=0.947317 time=51.4s + ttt_chunk [231/1904] bpb=0.945341 time=53.8s + ttt_chunk [241/1904] bpb=0.947721 time=56.1s + ttt_chunk [251/1904] bpb=0.946732 time=58.5s + ttt_chunk [261/1904] bpb=0.944192 time=60.9s + ttt_chunk [271/1904] bpb=0.944057 time=63.2s + ttt_chunk [281/1904] bpb=0.943612 time=65.6s + ttt_chunk [291/1904] bpb=0.942713 time=67.9s + ttt_chunk [301/1904] bpb=0.943475 time=70.3s + ttt_chunk [311/1904] bpb=0.944422 time=72.6s + ttt_chunk [321/1904] bpb=0.944382 time=75.0s + ttt_chunk [331/1904] bpb=0.944306 time=77.4s + ttt_chunk [341/1904] bpb=0.943192 time=79.7s + ttt_chunk [351/1904] bpb=0.942651 time=82.1s + ttt_chunk [361/1904] bpb=0.944115 time=84.5s + ttt_chunk [371/1904] bpb=0.943507 time=86.8s + ttt_chunk [381/1904] bpb=0.943059 time=89.2s + ttt_chunk [391/1904] bpb=0.942307 time=91.6s + ttt_chunk [401/1904] bpb=0.942080 time=93.9s + ttt_chunk [411/1904] bpb=0.941052 time=96.3s + ttt_chunk [421/1904] bpb=0.939784 time=98.7s + ttt_chunk [431/1904] bpb=0.938400 time=101.0s + ttt_chunk [441/1904] bpb=0.937732 time=103.4s + ttt_chunk [451/1904] bpb=0.935945 time=105.8s + ttt_chunk [461/1904] bpb=0.935694 time=108.2s + ttt_chunk [471/1904] bpb=0.934666 time=110.5s + ttt_chunk [481/1904] bpb=0.935068 time=112.9s + ttt_chunk [491/1904] bpb=0.935595 time=115.2s + ttt_chunk [501/1904] bpb=0.936116 time=117.6s + ttt_chunk [511/1904] bpb=0.936868 time=120.0s + ttt_chunk [521/1904] bpb=0.939505 time=122.3s + ttt_chunk [531/1904] bpb=0.939533 time=124.7s + ttt_chunk [541/1904] bpb=0.940639 time=127.1s + ttt_chunk [551/1904] bpb=0.940191 time=129.4s + ttt_chunk [561/1904] bpb=0.940047 time=131.8s + ttt_chunk [571/1904] bpb=0.940319 time=134.2s + ttt_chunk [581/1904] bpb=0.941310 time=136.5s + ttt_chunk [591/1904] bpb=0.941933 time=138.9s + ttt_chunk [601/1904] bpb=0.941911 time=141.3s + ttt_chunk [611/1904] bpb=0.941772 time=143.6s + ttt_chunk [621/1904] bpb=0.943185 time=146.0s + ttt_chunk [631/1904] bpb=0.943644 time=148.4s + ttt_chunk [641/1904] bpb=0.944237 time=150.7s + ttt_chunk [651/1904] bpb=0.943977 time=153.1s + ttt_chunk [661/1904] bpb=0.944482 time=155.4s + ttt_chunk [671/1904] bpb=0.945080 time=157.8s + ttt_chunk [681/1904] bpb=0.945138 time=160.2s + ttt_chunk [691/1904] bpb=0.945189 time=162.5s + ttt_chunk [701/1904] bpb=0.945827 time=164.9s + ttt_chunk [711/1904] bpb=0.946313 time=167.3s + ttt_chunk [721/1904] bpb=0.945798 time=169.5s + ttt_chunk [731/1904] bpb=0.945671 time=171.8s + ttt_chunk [741/1904] bpb=0.945600 time=174.1s + ttt_chunk [751/1904] bpb=0.945010 time=176.3s + ttt_chunk [761/1904] bpb=0.944838 time=178.6s + ttt_chunk [771/1904] bpb=0.944373 time=180.9s + ttt_chunk [781/1904] bpb=0.944208 time=183.3s + ttt_chunk [791/1904] bpb=0.943991 time=185.7s + ttt_chunk [801/1904] bpb=0.944312 time=188.0s + ttt_chunk [811/1904] bpb=0.945292 time=190.4s + ttt_chunk [821/1904] bpb=0.944678 time=192.8s + ttt_chunk [831/1904] bpb=0.945320 time=195.1s + ttt_chunk [841/1904] bpb=0.945218 time=197.5s + ttt_chunk [851/1904] bpb=0.945569 time=199.9s + ttt_chunk [861/1904] bpb=0.946017 time=202.3s + ttt_chunk [871/1904] bpb=0.946303 time=204.6s + ttt_chunk [881/1904] bpb=0.946467 time=207.0s + ttt_chunk [891/1904] bpb=0.947352 time=209.2s + ttt_chunk [901/1904] bpb=0.947989 time=211.6s + ttt_chunk [911/1904] bpb=0.948574 time=213.9s + ttt_chunk [921/1904] bpb=0.949399 time=216.3s + ttt_chunk [931/1904] bpb=0.949907 time=218.7s + ttt_chunk [941/1904] bpb=0.949558 time=221.0s + ttt_chunk [951/1904] bpb=0.949777 time=223.4s + ttt_chunk [961/1904] bpb=0.949834 time=225.7s + ttt_chunk [971/1904] bpb=0.950314 time=228.1s + ttt_chunk [981/1904] bpb=0.950653 time=230.4s + ttt_chunk [991/1904] bpb=0.950442 time=232.7s + ttt_chunk [1001/1904] bpb=0.950593 time=235.1s + ttt_chunk [1011/1904] bpb=0.951170 time=237.5s + ttt_chunk [1021/1904] bpb=0.951753 time=239.8s + ttt_chunk [1031/1904] bpb=0.951453 time=242.2s + ttt_chunk [1041/1904] bpb=0.951271 time=244.5s + ttt_chunk [1051/1904] bpb=0.950964 time=246.9s + ttt_chunk [1061/1904] bpb=0.950293 time=249.3s + ttt_chunk [1071/1904] bpb=0.950168 time=251.6s + ttt_chunk [1081/1904] bpb=0.950004 time=254.0s + ttt_chunk [1091/1904] bpb=0.949922 time=256.4s + ttt_chunk [1101/1904] bpb=0.950135 time=258.7s + ttt_chunk [1111/1904] bpb=0.950082 time=261.1s + ttt_chunk [1121/1904] bpb=0.950364 time=263.4s + ttt_chunk [1131/1904] bpb=0.950346 time=265.8s + ttt_chunk [1141/1904] bpb=0.950656 time=268.2s + ttt_chunk [1151/1904] bpb=0.950340 time=270.5s + ttt_chunk [1161/1904] bpb=0.950405 time=272.9s + ttt_chunk [1171/1904] bpb=0.950157 time=275.2s + ttt_chunk [1181/1904] bpb=0.950237 time=277.6s + ttt_chunk [1191/1904] bpb=0.949822 time=280.0s + ttt_chunk [1201/1904] bpb=0.950124 time=282.3s + ttt_chunk [1211/1904] bpb=0.950281 time=284.7s + ttt_chunk [1221/1904] bpb=0.950330 time=287.0s + ttt_chunk [1231/1904] bpb=0.951030 time=289.4s + ttt_chunk [1241/1904] bpb=0.951141 time=291.8s + ttt_chunk [1251/1904] bpb=0.951689 time=294.1s + ttt_chunk [1261/1904] bpb=0.951526 time=296.5s + ttt_chunk [1271/1904] bpb=0.951625 time=298.9s + ttt_chunk [1281/1904] bpb=0.951833 time=301.2s + ttt_chunk [1291/1904] bpb=0.952138 time=303.6s + ttt_chunk [1301/1904] bpb=0.952691 time=306.0s + ttt_chunk [1311/1904] bpb=0.952943 time=308.3s + ttt_chunk [1321/1904] bpb=0.953451 time=310.7s + ttt_chunk [1331/1904] bpb=0.953436 time=313.0s + ttt_chunk [1341/1904] bpb=0.953740 time=315.4s + ttt_chunk [1351/1904] bpb=0.954124 time=317.8s + ttt_chunk [1361/1904] bpb=0.954480 time=320.1s + ttt_chunk [1371/1904] bpb=0.954304 time=322.5s + ttt_chunk [1381/1904] bpb=0.953771 time=324.9s + ttt_chunk [1391/1904] bpb=0.953956 time=327.2s + ttt_chunk [1401/1904] bpb=0.953548 time=329.6s + ttt_chunk [1411/1904] bpb=0.953520 time=331.9s + ttt_chunk [1421/1904] bpb=0.953437 time=334.3s + ttt_chunk [1431/1904] bpb=0.953470 time=336.7s + ttt_chunk [1441/1904] bpb=0.953492 time=339.0s + ttt_chunk [1451/1904] bpb=0.953543 time=341.4s + ttt_chunk [1461/1904] bpb=0.953783 time=343.8s + ttt_chunk [1471/1904] bpb=0.953578 time=346.1s + ttt_chunk [1481/1904] bpb=0.953328 time=348.5s + ttt_chunk [1491/1904] bpb=0.952615 time=350.8s + ttt_chunk [1501/1904] bpb=0.953147 time=353.2s + ttt_chunk [1511/1904] bpb=0.953082 time=355.6s + ttt_chunk [1521/1904] bpb=0.952924 time=357.9s + ttt_chunk [1531/1904] bpb=0.952638 time=360.3s + ttt_chunk [1541/1904] bpb=0.952957 time=362.6s + ttt_chunk [1551/1904] bpb=0.953036 time=365.0s + ttt_chunk [1561/1904] bpb=0.953133 time=367.4s + ttt_chunk [1571/1904] bpb=0.953289 time=369.7s + ttt_chunk [1581/1904] bpb=0.953147 time=372.1s + ttt_chunk [1591/1904] bpb=0.952888 time=374.4s + ttt_chunk [1601/1904] bpb=0.953175 time=376.8s + ttt_chunk [1611/1904] bpb=0.953120 time=379.2s + ttt_chunk [1621/1904] bpb=0.953266 time=381.5s + ttt_chunk [1631/1904] bpb=0.953248 time=383.9s + ttt_chunk [1641/1904] bpb=0.953153 time=386.2s + ttt_chunk [1651/1904] bpb=0.953304 time=388.6s + ttt_chunk [1661/1904] bpb=0.953450 time=391.0s + ttt_chunk [1671/1904] bpb=0.953031 time=393.3s + ttt_chunk [1681/1904] bpb=0.953203 time=395.7s + ttt_chunk [1691/1904] bpb=0.953012 time=398.1s + ttt_chunk [1701/1904] bpb=0.953064 time=400.4s + ttt_chunk [1711/1904] bpb=0.953522 time=402.8s + ttt_chunk [1721/1904] bpb=0.953598 time=405.1s + ttt_chunk [1731/1904] bpb=0.954146 time=407.5s + ttt_chunk [1741/1904] bpb=0.954059 time=409.9s + ttt_chunk [1751/1904] bpb=0.954169 time=412.2s + ttt_chunk [1761/1904] bpb=0.954057 time=414.6s + ttt_chunk [1771/1904] bpb=0.954031 time=416.9s + ttt_chunk [1781/1904] bpb=0.953853 time=419.3s + ttt_chunk [1791/1904] bpb=0.953654 time=421.7s + ttt_chunk [1801/1904] bpb=0.953564 time=424.0s + ttt_chunk [1811/1904] bpb=0.953581 time=426.4s + ttt_chunk [1821/1904] bpb=0.953341 time=428.8s + ttt_chunk [1831/1904] bpb=0.953460 time=431.1s + ttt_chunk [1841/1904] bpb=0.953381 time=433.5s + ttt_chunk [1851/1904] bpb=0.953179 time=435.8s + ttt_chunk [1861/1904] bpb=0.953061 time=438.2s + ttt_chunk [1871/1904] bpb=0.952992 time=440.6s + ttt_chunk [1881/1904] bpb=0.953127 time=442.9s + ttt_chunk [1891/1904] bpb=0.952652 time=445.3s + ttt_chunk [1901/1904] bpb=0.952918 time=447.6s + ttt_chunk [1904/1904] bpb=0.952924 time=448.1s +ttt_sliding:done val_loss=1.928505 val_bpb=0.949127 elapsed=448.2s +legal_ttt val_loss:1.9285 val_bpb:0.9491 eval_time:448555ms +legal_ttt_exact val_loss:1.92850507 val_bpb:0.94912668 diff --git a/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed2025.log b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed2025.log new file mode 100644 index 0000000000..775d27aa2e --- /dev/null +++ b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed2025.log @@ -0,0 +1,2308 @@ +from __future__ import annotations +import copy +import glob +import io +import lzma +import math +import os +import random +import subprocess +import sys +import time +import uuid +from pathlib import Path +import numpy as np +import sentencepiece as spm +import torch +import torch.distributed as dist +import torch.nn.functional as F +from torch import Tensor, nn +from torch.nn.parallel import DistributedDataParallel as DDP +from flash_attn_interface import flash_attn_func as flash_attn_3_func +class Hyperparameters: + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp1024") + train_files = os.path.join(data_path, "fineweb_train_*.bin") + val_files = os.path.join(data_path, "fineweb_val_*.bin") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_1024_bpe.model") + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + seed = int(os.environ.get("SEED", 1337)) + val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 3500)) + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786_432)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 1.5)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 1024)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 3.0)) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.035)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.025)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.025)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.99)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + swa_enabled = bool(int(os.environ.get("SWA_ENABLED", "1"))) + swa_every = int(os.environ.get("SWA_EVERY", 50)) + muon_wd = float(os.environ.get("MUON_WD", 0.04)) + adam_wd = float(os.environ.get("ADAM_WD", 0.04)) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 2048)) + bigram_dim = int(os.environ.get("BIGRAM_DIM", 128)) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 4)) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + ve_enabled = bool(int(os.environ.get("VE_ENABLED", "1"))) + ve_dim = int(os.environ.get("VE_DIM", 128)) + ve_layers = os.environ.get("VE_LAYERS", "9,10") + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "0"))) + ttt_lr = float(os.environ.get("TTT_LR", 0.002)) + ttt_epochs = int(os.environ.get("TTT_EPOCHS", 3)) + ttt_chunk_tokens = int(os.environ.get("TTT_CHUNK_TOKENS", 32768)) + ttt_freeze_blocks = int(os.environ.get("TTT_FREEZE_BLOCKS", 2)) + ttt_momentum = float(os.environ.get("TTT_MOMENTUM", 0.9)) + ttt_batch_seqs = int(os.environ.get("TTT_BATCH_SEQS", 32)) + ttt_grad_clip = float(os.environ.get("TTT_GRAD_CLIP", 1.0)) + negative_slope = float(os.environ.get("NEGATIVE_SLOPE", 0.5)) + use_gptq = bool(int(os.environ.get("USE_GPTQ", "0"))) + gptq_calib_samples = int(os.environ.get("GPTQ_CALIB_SAMPLES", "64")) + gptq_reserve_ms = float(os.environ.get("GPTQ_RESERVE_MS", "14000")) + quant_clip_range = int(os.environ.get("QUANT_CLIP_RANGE", 31)) + tokenizer_meta_path = os.environ.get("TOKENIZER_META_PATH", "") + tokenizer_meta_validate = bool(int(os.environ.get("TOKENIZER_META_VALIDATE", "0"))) + +# --- Batched Newton-Schulz orthogonalization --- + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 5, eps: float = 1e-7) -> Tensor: + """Batched Newton-Schulz orthogonalization. G: (B,M,N) or (M,N).""" + a, b, c = (3.4445, -4.7750, 2.0315) + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + for _ in range(steps): + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + +# --- Parallel Muon optimizer --- + +class Muon(torch.optim.Optimizer): + """Parallel Muon: post-backward reduce-scatter -> local NS5 -> all-gather. + + No DDP for bank params. After backward, this optimizer: + 1. Launches async reduce-scatter for all banks (biggest first) + 2. Returns control so Adam can step on small params while RS is in-flight + 3. Waits for each RS, runs local NS5 on the shard, launches async all-gather + 4. Each all-gather overlaps with next bank's NS5 + """ + def __init__(self, params, lr: float, momentum: float, backend_steps: int, + nesterov: bool = True, weight_decay: float = 0.0): + super().__init__( + params, + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, + nesterov=nesterov, weight_decay=weight_decay), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + 'p': p, + 'B': B, + 'padded_grad': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard_mom': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'full_update': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'scale': max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + # Sort by size descending -- launch biggest reduce-scatters first + self._bank_meta.sort(key=lambda m: -m['p'].numel()) + self._built = True + + def launch_reduce_scatters(self): + """Phase 1: launch async reduce-scatter for all banks. Call right after backward.""" + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m['p'] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m['padded_grad'] + pg[:m['B']].copy_(p.grad.bfloat16()) + if pg.shape[0] > m['B']: + pg[m['B']:].zero_() + fut = dist.reduce_scatter_tensor(m['shard'], pg, op=dist.ReduceOp.AVG, async_op=True) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + """Phase 3: wait for RS, local NS5, all-gather. Call AFTER Adam steps.""" + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + + if not self._built: + self._build() + + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + + prev_ag_handle = None + prev_m = None + + sharded = self._distributed and hasattr(self, '_rs_futures') + + for i, m in enumerate(self._bank_meta): + p = m['p'] + if p.grad is None: + continue + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if sharded and self._rs_futures[i] is not None: + self._rs_futures[i].wait() + g = m['shard'] + buf = m['shard_mom'] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m['full_update'], update, async_op=True) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m['scale']) + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if hasattr(self, '_rs_futures'): + del self._rs_futures + + return loss + +# --- Tokenizer evaluation helpers --- + +def build_sentencepiece_luts( + sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device +) -> tuple[Tensor, Tensor, Tensor]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) + +TOKENIZER_META_FORMAT_VERSION = 1 +TOKENIZER_META_SUFFIX = ".meta.npz" + + +def _derive_tokenizer_meta_path(tokenizer_path: str) -> Path: + tokenizer = Path(tokenizer_path) + if tokenizer.suffix == ".model": + return tokenizer.with_suffix(TOKENIZER_META_SUFFIX) + return tokenizer.with_name(tokenizer.name + TOKENIZER_META_SUFFIX) + + +def build_sentencepiece_luts_np( + sp: spm.SentencePieceProcessor, vocab_size: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return base_bytes_np, has_leading_space_np, is_boundary_token_np + + +def load_tokenizer_meta_luts_np( + meta_path: Path, vocab_size: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray, dict[str, object]]: + def _scalar(value): + arr = np.asarray(value) + if arr.ndim == 0: + return arr.item() + first = arr.reshape(-1)[0] + return first.item() if hasattr(first, "item") else first + + with np.load(meta_path, allow_pickle=False) as data: + format_version = int(_scalar(data["format_version"])) + if format_version != TOKENIZER_META_FORMAT_VERSION: + raise ValueError( + f"Unsupported tokenizer meta format_version={format_version} " + f"expected={TOKENIZER_META_FORMAT_VERSION}" + ) + meta_vocab_size = int(_scalar(data["vocab_size"])) + tokenizer_kind = str(_scalar(data["tokenizer_kind"])) + source_model_name = str(_scalar(data["source_model_name"])) + base_bytes_np = np.asarray(data["base_bytes"], dtype=np.int16) + has_leading_space_np = np.asarray(data["has_leading_space"], dtype=np.bool_) + is_boundary_token_np = np.asarray(data["is_boundary_token"], dtype=np.bool_) + table_size = max(meta_vocab_size, vocab_size) + if base_bytes_np.shape[0] < table_size: + padded_base_bytes = np.zeros((table_size,), dtype=np.int16) + padded_has_leading_space = np.zeros((table_size,), dtype=np.bool_) + padded_is_boundary = np.ones((table_size,), dtype=np.bool_) + padded_base_bytes[: base_bytes_np.shape[0]] = base_bytes_np + padded_has_leading_space[: has_leading_space_np.shape[0]] = has_leading_space_np + padded_is_boundary[: is_boundary_token_np.shape[0]] = is_boundary_token_np + base_bytes_np = padded_base_bytes + has_leading_space_np = padded_has_leading_space + is_boundary_token_np = padded_is_boundary + metadata = { + "format_version": format_version, + "tokenizer_kind": tokenizer_kind, + "source_model_name": source_model_name, + "vocab_size": meta_vocab_size, + "meta_path": str(meta_path), + } + return base_bytes_np, has_leading_space_np, is_boundary_token_np, metadata + + +def load_tokenizer_luts( + tokenizer_path: str, + tokenizer_meta_path: str, + vocab_size: int, + device: torch.device, + *, + validate_meta: bool = False, +) -> tuple[tuple[Tensor, Tensor, Tensor], dict[str, object]]: + meta_path = ( + Path(tokenizer_meta_path) if tokenizer_meta_path + else _derive_tokenizer_meta_path(tokenizer_path) + ) + if meta_path.exists(): + base_bytes_np, has_leading_space_np, is_boundary_token_np, metadata = ( + load_tokenizer_meta_luts_np(meta_path, vocab_size) + ) + if validate_meta and str(tokenizer_path).endswith(".model"): + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + sp_luts = build_sentencepiece_luts_np(sp, vocab_size) + if not ( + np.array_equal(base_bytes_np, sp_luts[0]) + and np.array_equal(has_leading_space_np, sp_luts[1]) + and np.array_equal(is_boundary_token_np, sp_luts[2]) + ): + raise ValueError(f"Tokenizer metadata mismatch for {meta_path}") + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ), metadata + if not str(tokenizer_path).endswith(".model"): + raise FileNotFoundError(f"TOKENIZER_META_PATH does not exist: {meta_path}") + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + return build_sentencepiece_luts(sp, vocab_size, device), { + "tokenizer_kind": "sentencepiece", + "source_model_name": str(tokenizer_path), + "vocab_size": int(sp.vocab_size()), + "meta_path": None, + "fallback": True, + } + +def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = ((tokens.numel() - 1) // seq_len) * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] +def eval_val( + args: Hyperparameters, + model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + grad_accum_steps: int, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + seq_len = eval_seq_len or args.train_seq_len + local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + "VAL_BATCH_SIZE must provide at least one sequence per rank; " + f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " + f"GRAD_ACCUM_STEPS={grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + seq_start = (total_seqs * rank) // world_size + seq_end = (total_seqs * (rank + 1)) // world_size + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + model.eval() + with torch.inference_mode(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + batch_loss = model(x, y).detach() + batch_token_count = float(y.numel()) + val_loss_sum += batch_loss.to(torch.float64) * batch_token_count + val_token_count += batch_token_count + prev_ids = x.reshape(-1) + tgt_ids = y.reshape(-1) + token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + val_loss = val_loss_sum / val_token_count + bits_per_token = val_loss.item() / math.log(2.0) + tokens_per_byte = val_token_count.item() / val_byte_count.item() + model.train() + return float(val_loss.item()), float(bits_per_token * tokens_per_byte) + +# --- Quantization helpers --- + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,smear,dtg_gate,ve_layer_scales,ve_shared.scale,attn_gate,vr_lambda", + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_FP32_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "INT8_KEEP_FLOAT_FP32_NAME_PATTERNS", + ",".join(CONTROL_TENSOR_NAME_PATTERNS), + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_MAX_NUMEL = 65_536 +INT8_KEEP_FLOAT_STORE_DTYPE = torch.float16 +INT8_PER_ROW_SCALE_DTYPE = torch.float16 +INT8_CLIP_PERCENTILE = 99.99984 +INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 +def tensor_nbytes(t: Tensor) -> int: + return int(t.numel()) * int(t.element_size()) +def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: + if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): + return t.float().contiguous() + if t.dtype in {torch.float32, torch.bfloat16}: + passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") + return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() + return t +def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + clip_abs = ( + torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) + if t32.numel() + else torch.empty((t32.shape[0],), dtype=torch.float32) + ) + clipped = torch.maximum(torch.minimum(t32, clip_abs[:, None]), -clip_abs[:, None]) + scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) + q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() + return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() + clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 + scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) + q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() + return q, scale +def quantize_state_dict_int8(state_dict: dict[str, Tensor]): + quantized: dict[str, Tensor] = {} + scales: dict[str, Tensor] = {} + dtypes: dict[str, str] = {} + passthrough: dict[str, Tensor] = {} + passthrough_orig_dtypes: dict[str, str] = {} + qmeta: dict[str, dict[str, object]] = {} + stats = dict.fromkeys( + ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), + 0, + ) + for name, tensor in state_dict.items(): + t = tensor.detach().to("cpu").contiguous() + stats["param_count"] += int(t.numel()) + stats["num_tensors"] += 1 + stats["baseline_tensor_bytes"] += tensor_nbytes(t) + if not t.is_floating_point(): + stats["num_nonfloat_tensors"] += 1 + passthrough[name] = t + stats["int8_payload_bytes"] += tensor_nbytes(t) + continue + if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: + kept = keep_float_tensor(name, t, passthrough_orig_dtypes) + passthrough[name] = kept + stats["int8_payload_bytes"] += tensor_nbytes(kept) + continue + stats["num_float_tensors"] += 1 + q, s = quantize_float_tensor(t) + if s.ndim > 0: + qmeta[name] = {"scheme": "per_row", "axis": 0} + quantized[name] = q + scales[name] = s + dtypes[name] = str(t.dtype).removeprefix("torch.") + stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) + obj: dict[str, object] = { + "__quant_format__": "int8_clean_per_row_v1", + "quantized": quantized, + "scales": scales, + "dtypes": dtypes, + "passthrough": passthrough, + } + if qmeta: + obj["qmeta"] = qmeta + if passthrough_orig_dtypes: + obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes + return obj, stats +def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + qmeta = obj.get("qmeta", {}) + passthrough_orig_dtypes = obj.get("passthrough_orig_dtypes", {}) + for name, q in obj["quantized"].items(): + dtype = getattr(torch, obj["dtypes"][name]) + s = obj["scales"][name] + if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: + s = s.to(dtype=torch.float32) + out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() + else: + scale = float(s.item()) + out[name] = (q.float() * scale).to(dtype=dtype).contiguous() + for name, t in obj["passthrough"].items(): + out_t = t.detach().to("cpu").contiguous() + orig_dtype = passthrough_orig_dtypes.get(name) + if isinstance(orig_dtype, str): + out_t = out_t.to(dtype=getattr(torch, orig_dtype)).contiguous() + out[name] = out_t + return out + +# --- Data loading --- + +def load_data_shard(file: Path) -> Tensor: + header_bytes = 256 * np.dtype(" int: + key = str(file) + cached = _SHARD_NTOKENS_CACHE.get(key) + if cached is not None: + return cached + header = np.fromfile(file, dtype=" np.memmap: + key = str(file) + mm = _MMAP_CACHE.get(key) + if mm is not None: + return mm + n = _read_num_tokens(file) + mm = np.memmap(file, mode="r", dtype=" int: + if n <= 1: + return 1 + while True: + s = int(self._rng.integers(1, n)) + if math.gcd(s, n) == 1: + return s + def _reset_cursor(self, si: int, seq_len: int) -> None: + nt = int(self._num_tokens[si]) + max_phase = min(seq_len - 1, max(0, nt - seq_len - 1)) + phase = int(self._rng.integers(max_phase + 1)) if max_phase > 0 else 0 + bc = (nt - 1 - phase) // seq_len + self._cursor_phase[si] = phase + self._cursor_block_count[si] = bc + self._cursor_next[si] = 0 + self._cursor_start[si] = int(self._rng.integers(bc)) if bc > 1 else 0 + self._cursor_stride[si] = self._pick_coprime_stride(bc) + self._cursor_init[si] = True + def _ensure_cursor(self, si: int, seq_len: int) -> None: + if not self._cursor_init[si] or self._cursor_next[si] >= self._cursor_block_count[si]: + self._reset_cursor(si, seq_len) + def _take_from_shard(self, si: int, seq_len: int, count: int, out: list[tuple[int, int]]) -> None: + rem = count + while rem > 0: + self._ensure_cursor(si, seq_len) + bc = int(self._cursor_block_count[si]) + ni = int(self._cursor_next[si]) + take = min(rem, bc - ni) + phase = int(self._cursor_phase[si]) + start = int(self._cursor_start[si]) + stride = int(self._cursor_stride[si]) + for j in range(take): + bi = (start + (ni + j) * stride) % bc + out.append((si, phase + bi * seq_len)) + self._cursor_next[si] = ni + take + rem -= take + def _init_pipeline(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> None: + local_tokens = global_tokens // (self.world_size * grad_accum_steps) + num_seqs = local_tokens // seq_len + global_num_seqs = num_seqs * self.world_size + self._cfg = (local_tokens, seq_len, num_seqs, global_num_seqs) + bbc = (self._num_tokens - 1) // seq_len + eligible = bbc > 0 + self._eligible_shards = np.nonzero(eligible)[0].astype(np.int64) + self._base_block_counts = bbc[self._eligible_shards].astype(np.int64) + def _sample_global_windows(self) -> list[tuple[int, int]]: + assert self._cfg is not None and self._eligible_shards is not None + _, seq_len, _, gns = self._cfg + ec = int(self._eligible_shards.size) + progress = min(self._batches_built / 1800.0, 1.0) + remaining = np.empty(ec, dtype=np.float64) + for i, si in enumerate(self._eligible_shards.tolist()): + if self._cursor_init[si]: + r = int(self._cursor_block_count[si]) - int(self._cursor_next[si]) + remaining[i] = float(max(r, 1)) + else: + remaining[i] = float(self._base_block_counts[i]) + alpha = 0.90 - 0.40 * progress + weights = np.power(remaining, alpha) + ws = float(weights.sum()) + if not np.isfinite(ws) or ws <= 0.0: + weights = np.ones(ec, dtype=np.float64) + ws = float(weights.sum()) + probs = weights / ws + low = min(max(8, self.world_size), ec, gns) + high = min(max(32, self.world_size * 8), ec, gns) + mix = max(1, min(int(round(low + progress * (high - low))), ec, gns)) + cp = self._rng.choice(ec, size=mix, replace=False, p=probs) + cs = self._eligible_shards[cp] + cpr = probs[cp].copy() + cpr /= cpr.sum() + counts = np.ones(mix, dtype=np.int64) + extra = gns - mix + if extra > 0: + counts += self._rng.multinomial(extra, cpr).astype(np.int64) + perm = self._rng.permutation(mix) + cs, counts = cs[perm], counts[perm] + buckets: list[list[tuple[int, int]]] = [] + for si, cnt in zip(cs.tolist(), counts.tolist()): + b: list[tuple[int, int]] = [] + self._take_from_shard(int(si), seq_len, int(cnt), b) + if b: + if len(b) > 1: + bp = self._rng.permutation(len(b)) + b = [b[int(k)] for k in bp.tolist()] + buckets.append(b) + windows: list[tuple[int, int]] = [] + active = [i for i, bk in enumerate(buckets) if bk] + while active: + order = self._rng.permutation(len(active)) + new_active: list[int] = [] + for oi in order.tolist(): + bi = active[oi] + if buckets[bi]: + windows.append(buckets[bi].pop()) + if buckets[bi]: + new_active.append(bi) + active = new_active + return windows + def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: + if self._cfg is None: + self._init_pipeline(global_tokens, seq_len, grad_accum_steps) + _, _, num_seqs, gns = self._cfg + gw = self._sample_global_windows() + local_w = gw[self.rank::self.world_size] + x = torch.empty((num_seqs, seq_len), dtype=torch.int64) + y = torch.empty((num_seqs, seq_len), dtype=torch.int64) + for slot, (si, pos) in enumerate(local_w): + mm = _get_shard_memmap(self.files[si]) + window = torch.as_tensor(np.array(mm[pos:pos + seq_len + 1], dtype=np.int64)) + x[slot] = window[:-1] + y[slot] = window[1:] + self._batches_built += 1 + return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) + +# --- Transformer modules --- + +class RMSNorm(nn.Module): + def __init__(self, eps: float | None = None): + super().__init__() + self.eps = eps + def forward(self, x: Tensor) -> Tensor: + return F.rms_norm(x, (x.size(-1),), eps=self.eps) +class CastedLinear(nn.Linear): + def forward(self, x: Tensor) -> Tensor: + w = self.weight.to(x.dtype) + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) +def restore_low_dim_params_to_fp32(module: nn.Module) -> None: + with torch.no_grad(): + for name, param in module.named_parameters(): + if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: + param.data = param.data.float() +class Rotary(nn.Module): + def __init__(self, dim: int, base: float = 10000.0, train_seq_len: int = 1024, rope_dims: int = 0): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / (base ** (torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached: Tensor | None = None + self._sin_cached: Tensor | None = None + def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached != seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * (scale ** (rd / (rd - 2))) + inv_freq = 1.0 / (new_base ** (torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd)) + else: + inv_freq = self.inv_freq.to(device) + t = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor, rope_dims: int = 0) -> Tensor: + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + +class CausalSelfAttention(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + rope_base: float, + qk_gain_init: float, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + # No CastedLinear -- weights come from banks + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 # set by GPT.__init__ for partial RoPE + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False # set by GPT.__init__ for deep layers only + def _xsa_efficient(self, y: Tensor, v: Tensor) -> Tensor: + """Efficient XSA: subtract self-value projection via GQA-aware reshape (no repeat_interleave). + y: [B, T, H, D], v: [B, T, Hkv, D]. H must be divisible by Hkv.""" + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) # [B, T, Hkv, group, D] + vn = F.normalize(v, dim=-1).unsqueeze(-2) # [B, T, Hkv, 1, D] -- broadcast ready + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, v_embed: Tensor | None = None) -> Tensor: + if getattr(self, '_save_gptq', False): + self._gptq_qkv_in = x.detach() + bsz, seqlen, dim = x.shape + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, v_w.to(x.dtype)) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + y = flash_attn_3_func(q, k, v, causal=True) + if self.use_xsa: + y = self._xsa_efficient(y, v) + y = y.reshape(bsz, seqlen, dim) + if getattr(self, '_save_gptq', False): + self._gptq_o_in = y.detach() + return F.linear(y, out_w.to(x.dtype)) + +class SmearGate(nn.Module): + def __init__(self, dim: int): + super().__init__() + self.gate = nn.Parameter(torch.zeros(dim, dtype=torch.float32)) + def forward(self, x: Tensor) -> Tensor: + g = torch.sigmoid(self.gate.to(dtype=x.dtype))[None, None, :] + x_prev = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) + return (1 - g) * x + g * x_prev + +class BigramHashEmbedding(nn.Module): + def __init__(self, bigram_vocab_size: int, bigram_dim: int, model_dim: int): + super().__init__() + self.bigram_vocab_size = bigram_vocab_size + self.embed = nn.Embedding(bigram_vocab_size, bigram_dim) + nn.init.zeros_(self.embed.weight) + self.proj = CastedLinear(bigram_dim, model_dim, bias=False) if bigram_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.05, dtype=torch.float32)) + def bigram_hash(self, tokens: Tensor) -> Tensor: + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., 0] = mod + out[..., 1:] = torch.bitwise_xor(36313 * t[..., 1:], 27191 * t[..., :-1]) % mod + return out.long() + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(self.bigram_hash(token_ids)) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class ValueEmbedding(nn.Module): + """Reinject token identity into attention values at specific layers. + Each table maps vocab tokens to a low-dim embedding, projected to model_dim.""" + def __init__(self, vocab_size: int, ve_dim: int, model_dim: int): + super().__init__() + self.embed = nn.Embedding(vocab_size, ve_dim) + nn.init.normal_(self.embed.weight, std=0.01) + self.proj = CastedLinear(ve_dim, model_dim, bias=False) if ve_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.1, dtype=torch.float32)) + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(token_ids) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class MLP(nn.Module): + def __init__(self, dim: int, mlp_mult: int, neg_slope: float = 0.5): + super().__init__() + self.neg_slope = neg_slope + # No CastedLinear -- weights come from banks + def forward(self, x: Tensor, up_w: Tensor, down_w: Tensor) -> Tensor: + if getattr(self, '_save_gptq', False): + self._gptq_up_in = x.detach() + x = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=self.neg_slope) + x2 = x.square() + if getattr(self, '_save_gptq', False): + self._gptq_down_in = x2.detach() + return F.linear(x2, down_w.to(x.dtype)) + +class Block(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + rope_base: float, + qk_gain_init: float, + layer_idx: int = 0, + ln_scale: bool = False, + neg_slope: float = 0.5, + ): + super().__init__() + self.layer_idx = layer_idx + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = MLP(dim, mlp_mult, neg_slope=neg_slope) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + def forward(self, x: Tensor, x0: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, up_w: Tensor, down_w: Tensor, v_embed: Tensor | None = None) -> Tensor: + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, q_w, k_w, v_w, out_w, v_embed=v_embed) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + mlp_out = self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + return x_out + mlp_out + +class GPT(nn.Module): + def __init__( + self, + vocab_size: int, + num_layers: int, + model_dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + tie_embeddings: bool, + tied_embed_init_std: float, + logit_softcap: float, + rope_base: float, + qk_gain_init: float, + bigram_vocab_size: int = 0, + bigram_dim: int = 128, + xsa_last_n: int = 0, + rope_dims: int = 0, + ln_scale: bool = False, + ve_enabled: bool = False, + ve_dim: int = 128, + ve_layers: str = "9,10", + neg_slope: float = 0.5, + ): + super().__init__() + self._ve_target_dim = num_kv_heads * (model_dim // num_heads) # kv_dim for value projection + if logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") + self.tie_embeddings = tie_embeddings + self.tied_embed_init_std = tied_embed_init_std + self.logit_softcap = logit_softcap + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + # Parameter banks: contiguous 3D tensors for batched optimizer + head_dim = model_dim // num_heads + kv_dim = num_kv_heads * head_dim + mlp_dim = int(mlp_mult * model_dim) + self.num_layers = num_layers + self.qo_bank = nn.Parameter(torch.empty(2 * num_layers, model_dim, model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * num_layers, kv_dim, model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(num_layers, mlp_dim, model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(num_layers, model_dim, mlp_dim)) + self.blocks = nn.ModuleList( + [ + Block( + model_dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + layer_idx=i, + ln_scale=ln_scale, + neg_slope=neg_slope, + ) + for i in range(num_layers) + ] + ) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + kv_dim_ve = self._ve_target_dim + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim_ve) + self.ve_layer_scales = nn.ParameterList( + [nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices] + ) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.value_embeds = nn.ModuleList() # keep empty for compat + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + if self.lm_head is not None: + self.lm_head._zero_init = True + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + self._init_weights() + def _init_weights(self) -> None: + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + # Init banks: orthogonal, with proj layers scaled down and out/down zero-init + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) # Q + nn.init.zeros_(self.qo_bank.data[n + i]) # Out (zero init) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) # K + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) # V + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) # MLP up + nn.init.zeros_(self.mlp_down_bank.data[i]) # MLP down (zero init) + # Scale proj layers (out_proj and mlp_down are "proj" layers) + self.qo_bank.data[n + i].mul_(proj_scale) + self.mlp_down_bank.data[i].mul_(proj_scale) + # Init remaining nn.Linear modules (bigram proj, lm_head) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif module.weight.ndim == 2 and module.weight.shape[0] >= 64 and module.weight.shape[1] >= 64: + nn.init.orthogonal_(module.weight, gain=1.0) + def _get_ve(self, layer_idx: int, input_ids: Tensor, ve_cache: dict | None = None) -> Tensor | None: + """Get value embedding for a specific layer using shared table + per-layer scale.""" + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if ve_cache is not None and 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_base = ve_cache['ve'] if ve_cache is not None else self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_base * self.ve_layer_scales[ve_idx].to(dtype=ve_base.dtype) + def forward(self, input_ids: Tensor, target_ids: Tensor) -> Tensor: + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + if self.tie_embeddings: + logits_proj = F.linear(x_flat, self.tok_emb.weight) + else: + if self.lm_head is None: + raise RuntimeError("lm_head is required when tie_embeddings=False") + logits_proj = self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy(logits.float(), targets, reduction="mean") + def forward_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve) + x = self.final_norm(x) + if self.tie_embeddings: + logits_proj = F.linear(x, self.tok_emb.weight) + else: + logits_proj = self.lm_head(x) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + +# --- Sliding window evaluation --- + +def eval_val_sliding( + args: Hyperparameters, + base_model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + stride: int, + batch_seqs: int = 32, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + """Sliding window evaluation: each token scored with maximum context.""" + seq_len = eval_seq_len or args.train_seq_len + total_tokens = val_tokens.numel() - 1 + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= 1] + total_windows = len(window_starts) + my_s = (total_windows * rank) // world_size + my_e = (total_windows * (rank + 1)) // world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + base_model.eval() + compiled_logits = torch.compile(base_model.forward_logits, dynamic=False, fullgraph=True) + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk[:-1] + y_batch[i, :wlen] = chunk[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = compiled_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), + reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt = y_batch[i, s:wlen] + prev = x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + val_loss = (loss_sum / token_count).item() + bits_per_token = val_loss / math.log(2.0) + tokens_per_byte = token_count.item() / byte_count.item() + base_model.train() + return val_loss, bits_per_token * tokens_per_byte + + +def eval_val_sliding_ttt( + args: Hyperparameters, base_model: nn.Module, rank: int, world_size: int, + device: torch.device, val_tokens: Tensor, base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, is_boundary_token_lut: Tensor, + stride: int, batch_seqs: int = 32, log0=print, +) -> tuple[float, float]: + """Legal score-first TTT (PR #461 recipe): score each chunk with sliding windows, + then train on it. Every token scored BEFORE any update that could use it.""" + seq_len = args.train_seq_len + total_tokens = val_tokens.numel() - 1 + ttt_chunk = args.ttt_chunk_tokens + + # Pre-compute all window starts + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= stride or ws == 0] + + # Assign each window to a chunk based on the first token it scores + num_chunks = (total_tokens + ttt_chunk - 1) // ttt_chunk + chunk_windows: list[list[int]] = [[] for _ in range(num_chunks)] + for ws in window_starts: + end = min(ws + seq_len, total_tokens) + wlen = end - ws + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_start = ws + s + ci = min(scored_start // ttt_chunk, num_chunks - 1) + chunk_windows[ci].append(ws) + + log0(f"ttt_sliding:start chunks={num_chunks} chunk_tokens={ttt_chunk} " + f"total_windows={len(window_starts)} stride={stride} " + f"ttt_lr={args.ttt_lr} ttt_epochs={args.ttt_epochs} " + f"freeze_blocks={args.ttt_freeze_blocks}") + + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + + # Freeze first N blocks + frozen_block_ids = set(range(min(args.ttt_freeze_blocks, len(base_model.blocks)))) + ttt_params = [] + for name, p in base_model.named_parameters(): + freeze = False + for bi in frozen_block_ids: + if f"blocks.{bi}." in name: + freeze = True + break + if freeze: + p.requires_grad_(False) + else: + p.requires_grad_(True) + ttt_params.append(p) + + log0(f"ttt_sliding:params unfrozen={sum(p.numel() for p in ttt_params)} " + f"frozen={sum(p.numel() for p in base_model.parameters() if not p.requires_grad)}") + + optimizer = torch.optim.SGD(ttt_params, lr=args.ttt_lr, momentum=args.ttt_momentum) + t0 = time.perf_counter() + + for ci in range(num_chunks): + windows = chunk_windows[ci] + if not windows: + continue + chunk_start = ci * ttt_chunk + chunk_end = min((ci + 1) * ttt_chunk, total_tokens) + + # --- Phase 1: SCORE this chunk's windows (inference_mode) --- + my_s = (len(windows) * rank) // world_size + my_e = (len(windows) * (rank + 1)) // world_size + my_windows = windows[my_s:my_e] + + base_model.eval() + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk_tok = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk_tok[:-1] + y_batch[i, :wlen] = chunk_tok[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = base_model.forward_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt, prev = y_batch[i, s:wlen], x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + + # --- Phase 2: TRAIN on this chunk (already scored = legal) --- + is_last_chunk = (ci == num_chunks - 1) + if not is_last_chunk and args.ttt_epochs > 0: + base_model.train() + chunk_seqs = (chunk_end - chunk_start) // seq_len + if chunk_seqs > 0: + cos_lr = args.ttt_lr * 0.5 * (1.0 + math.cos(math.pi * ci / max(num_chunks - 1, 1))) + for pg in optimizer.param_groups: + pg['lr'] = cos_lr + my_seq_s = (chunk_seqs * rank) // world_size + my_seq_e = (chunk_seqs * (rank + 1)) // world_size + my_chunk_seqs = my_seq_e - my_seq_s + for _ep in range(args.ttt_epochs): + for bs in range(0, my_chunk_seqs, args.ttt_batch_seqs): + be = min(bs + args.ttt_batch_seqs, my_chunk_seqs) + actual_bs = my_seq_s + bs + start_tok = chunk_start + actual_bs * seq_len + end_tok = chunk_start + (my_seq_s + be) * seq_len + 1 + if end_tok > val_tokens.numel(): + continue + local = val_tokens[start_tok:end_tok].to(device=device, dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + optimizer.zero_grad(set_to_none=True) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + loss = base_model(x, y) + loss.backward() + if world_size > 1: + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + torch.nn.utils.clip_grad_norm_(ttt_params, args.ttt_grad_clip) + optimizer.step() + + if rank == 0 and (ci % 10 == 0 or ci == num_chunks - 1): + elapsed = time.perf_counter() - t0 + rl = loss_sum.item() / max(token_count.item(), 1) + rbpb = rl / math.log(2.0) * (token_count.item() / max(byte_count.item(), 1)) if token_count.item() > 0 else 0.0 + log0(f" ttt_chunk [{ci+1}/{num_chunks}] bpb={rbpb:.6f} time={elapsed:.1f}s") + + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_count.item()) + + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + + log0(f"ttt_sliding:done val_loss={val_loss:.6f} val_bpb={val_bpb:.6f} " + f"elapsed={time.perf_counter() - t0:.1f}s") + return val_loss, val_bpb + + +# --- GPTQ-lite int6 quantization --- + +def _classify_param(name: str) -> str: + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or (".proj." in name and ".mlp." not in name): + return "attn" + return "other" +def quantize_int6_per_row(t: Tensor, clip_range: int = 31) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + best_q, best_s, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(t32.abs(), pct, dim=1) + else: + row_clip = t32.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + recon = q.float() * s.float()[:, None] + err = (t32 - recon).pow(2).mean().item() + if err < best_err: + best_q, best_s, best_err = q, s, err + return best_q, best_s + amax = t32.abs().max().item() + scale = torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + q = torch.clamp(torch.round(t32 / scale.float()), -clip_range, clip_range).to(torch.int8) + return q, scale + +def _unbank_state_dict(sd: dict[str, Tensor], num_layers: int) -> dict[str, Tensor]: + """Convert 3D bank tensors into individual 2D tensors with standard names.""" + out: dict[str, Tensor] = {} + n = num_layers + for name, tensor in sd.items(): + if name == "qo_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_q.weight"] = tensor[i] + out[f"blocks.{i}.attn.proj.weight"] = tensor[n + i] + elif name == "kv_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_k.weight"] = tensor[i] + out[f"blocks.{i}.attn.c_v.weight"] = tensor[n + i] + elif name == "mlp_up_bank": + for i in range(n): + out[f"blocks.{i}.mlp.fc.weight"] = tensor[i] + elif name == "mlp_down_bank": + for i in range(n): + out[f"blocks.{i}.mlp.proj.weight"] = tensor[i] + else: + out[name] = tensor + return out + +def _rebank_state_dict(sd: dict[str, Tensor], num_layers: int, template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + """Convert individual 2D tensors back into 3D bank tensors.""" + out: dict[str, Tensor] = {} + n = num_layers + # Reconstruct banks from individual weight keys + qo_slices = [None] * (2 * n) + kv_slices = [None] * (2 * n) + up_slices = [None] * n + down_slices = [None] * n + consumed = set() + for i in range(n): + qk = f"blocks.{i}.attn.c_q.weight" + if qk in sd: + qo_slices[i] = sd[qk] + consumed.add(qk) + ok = f"blocks.{i}.attn.proj.weight" + if ok in sd: + qo_slices[n + i] = sd[ok] + consumed.add(ok) + kk = f"blocks.{i}.attn.c_k.weight" + if kk in sd: + kv_slices[i] = sd[kk] + consumed.add(kk) + vk = f"blocks.{i}.attn.c_v.weight" + if vk in sd: + kv_slices[n + i] = sd[vk] + consumed.add(vk) + fk = f"blocks.{i}.mlp.fc.weight" + if fk in sd: + up_slices[i] = sd[fk] + consumed.add(fk) + dk = f"blocks.{i}.mlp.proj.weight" + if dk in sd: + down_slices[i] = sd[dk] + consumed.add(dk) + out["qo_bank"] = torch.stack(qo_slices).to(dtype=template_sd["qo_bank"].dtype) + out["kv_bank"] = torch.stack(kv_slices).to(dtype=template_sd["kv_bank"].dtype) + out["mlp_up_bank"] = torch.stack(up_slices).to(dtype=template_sd["mlp_up_bank"].dtype) + out["mlp_down_bank"] = torch.stack(down_slices).to(dtype=template_sd["mlp_down_bank"].dtype) + for name, tensor in sd.items(): + if name not in consumed: + out[name] = tensor + return out + +def mixed_quantize_int6(state_dict: dict[str, Tensor], int6_cats: set[str], clip_range: int = 31, + hessians: dict[str, Tensor] | None = None): + num_layers_total = max( + (int(k.split(".")[1]) for k in state_dict if k.startswith("blocks.")), + default=0, + ) + 1 + late_k_layers = set(range(num_layers_total - 2, num_layers_total)) + result: dict[str, Tensor] = {} + meta: dict[str, object] = {} + gptq_count, naive_count = 0, 0 + for name, tensor in state_dict.items(): + t = tensor.detach().cpu().contiguous() + cat = _classify_param(name) + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough" + continue + if any(p in name for p in CONTROL_TENSOR_NAME_PATTERNS): + result[name] = t.float() + meta[name] = "passthrough_ctrl" + continue + if cat in int6_cats and t.ndim >= 1: + H = hessians.get(name) if hessians else None + if H is not None and t.ndim == 2: + q, s = gptq_quantize_weight(t, H.cpu(), clip_range=clip_range) + gptq_count += 1 + else: + q, s = quantize_int6_per_row(t, clip_range=clip_range) + naive_count += 1 + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int6"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + if hessians: + print(f"gptq_quantize: {gptq_count} GPTQ layers, {naive_count} naive layers", flush=True) + return result, meta +def dequantize_mixed_int6(result: dict[str, Tensor], meta: dict[str, object], + template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + for name, orig in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if info in ("passthrough", "passthrough_ctrl", "passthrough_fp16"): + t = result[name] + if t.dtype == torch.float16 and orig_dtype in (torch.float32, torch.bfloat16): + t = t.to(orig_dtype) + out[name] = t + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out + +# --- Full Hessian GPTQ --- + +def gptq_quantize_weight(W: Tensor, H: Tensor, clip_range: int = 31, + block_size: int = 128, percdamp: float = 0.01) -> tuple[Tensor, Tensor]: + """GPTQ with Cholesky error compensation and actorder (Frantar et al., ICLR 2023).""" + W_orig = W.float().clone() + rows, cols = W_orig.shape + H = H.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = percdamp * H.diag().mean() + H.diagonal().add_(damp) + perm = torch.argsort(H.diag(), descending=True) + invperm = torch.argsort(perm) + W_perm = W_orig[:, perm].clone() + W_perm[:, dead[perm]] = 0 + H = H[perm][:, perm] + try: + Hinv = torch.cholesky_inverse(torch.linalg.cholesky(H)) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + except torch.linalg.LinAlgError: + return quantize_int6_per_row(W_orig, clip_range) + best_q, best_scale, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(W_orig.abs(), pct, dim=1) + else: + row_clip = W_orig.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + sf = s.float() + Q = torch.zeros(rows, cols, dtype=torch.int8) + W_work = W_perm.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + W_block = W_work[:, i1:i2].clone() + Hinv_block = Hinv[i1:i2, i1:i2] + Err = torch.zeros(rows, i2 - i1) + for j in range(i2 - i1): + w_col = W_block[:, j] + d = Hinv_block[j, j] + q_col = torch.clamp(torch.round(w_col / sf), -clip_range, clip_range) + Q[:, i1 + j] = q_col.to(torch.int8) + err = (w_col - q_col.float() * sf) / d + Err[:, j] = err + W_block[:, j:] -= err.unsqueeze(1) * Hinv_block[j, j:].unsqueeze(0) + if i2 < cols: + W_work[:, i2:] -= Err @ Hinv[i1:i2, i2:] + recon = Q.float() * sf[:, None] + mse = (W_perm - recon).pow(2).mean().item() + if mse < best_err: + best_q, best_scale, best_err = Q, s, mse + best_q = best_q[:, invperm] + return best_q, best_scale + +def _init_hessians(nl: int, dim: int, mlp_dim: int, device: torch.device) -> dict[str, Tensor]: + h: dict[str, Tensor] = {} + for i in range(nl): + for k in ['c_q', 'c_k', 'c_v']: + h[f'blocks.{i}.attn.{k}.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.attn.proj.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.mlp.fc.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.mlp.proj.weight'] = torch.zeros(mlp_dim, mlp_dim, dtype=torch.float32, device=device) + return h + +def _accum_hessians(hessians: dict[str, Tensor], blocks: nn.ModuleList, dim: int, mlp_dim: int) -> None: + for i, block in enumerate(blocks): + qkv_in = block.attn._gptq_qkv_in.float().reshape(-1, dim) + h_qkv = qkv_in.t() @ qkv_in + hessians[f'blocks.{i}.attn.c_q.weight'] += h_qkv + hessians[f'blocks.{i}.attn.c_k.weight'] += h_qkv + hessians[f'blocks.{i}.attn.c_v.weight'] += h_qkv + o_in = block.attn._gptq_o_in.float().reshape(-1, dim) + hessians[f'blocks.{i}.attn.proj.weight'] += o_in.t() @ o_in + up_in = block.mlp._gptq_up_in.float().reshape(-1, dim) + hessians[f'blocks.{i}.mlp.fc.weight'] += up_in.t() @ up_in + down_in = block.mlp._gptq_down_in.float().reshape(-1, mlp_dim) + hessians[f'blocks.{i}.mlp.proj.weight'] += down_in.t() @ down_in + +def _finalize_hessians(hessians: dict[str, Tensor], num_batches: int) -> None: + for name in hessians: + hessians[name] = hessians[name].cpu() / num_batches + damp = 0.01 * torch.diag(hessians[name]).mean().clamp_min(1e-6) + hessians[name] += damp * torch.eye(hessians[name].shape[0]) + +def gptq_collect_hessians(base_model: nn.Module, train_loader, device: torch.device, + num_batches: int, batch_tokens: int, seq_len: int, + grad_accum_steps: int) -> dict[str, Tensor]: + """Collect Hessians H = X^T X from training data.""" + nl = base_model.num_layers + dim = base_model.tok_emb.weight.shape[1] + mlp_dim = base_model.mlp_up_bank.shape[1] + hessians = _init_hessians(nl, dim, mlp_dim, device) + for block in base_model.blocks: + block.attn._save_gptq = True + block.mlp._save_gptq = True + base_model.eval() + with torch.inference_mode(), torch.autocast(device_type='cuda', dtype=torch.bfloat16): + for _ in range(num_batches): + x, y = train_loader.next_batch(batch_tokens, seq_len, grad_accum_steps) + base_model(x, y) + _accum_hessians(hessians, base_model.blocks, dim, mlp_dim) + for block in base_model.blocks: + block.attn._save_gptq = False + block.mlp._save_gptq = False + _finalize_hessians(hessians, num_batches) + base_model.train() + return hessians + +# --- Training --- + +def main() -> None: + code = Path(__file__).read_text(encoding="utf-8") + args = Hyperparameters() + # zeropower_via_newtonschulz5 runs eagerly with bmm -- do NOT compile + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError(f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral") + grad_accum_steps = 8 // world_size + grad_scale = 1.0 / grad_accum_steps + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + master_process = rank == 0 + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + logfile = None + if master_process: + os.makedirs("logs", exist_ok=True) + logfile = f"logs/{args.run_id}.txt" + print(logfile) + def log0(msg: str, console: bool = True) -> None: + if not master_process: + return + if console: + print(msg) + if logfile is not None: + with open(logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + log0(code, console=False) + log0("=" * 100, console=False) + log0(f"Running Python {sys.version}", console=False) + log0(f"Running PyTorch {torch.__version__}", console=False) + log0( + subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=False).stdout, + console=False, + ) + log0("=" * 100, console=False) + random.seed(args.seed) + np.random.seed(args.seed) + torch.manual_seed(args.seed) + torch.cuda.manual_seed_all(args.seed) + (base_bytes_lut, has_leading_space_lut, is_boundary_token_lut), tokenizer_metadata = load_tokenizer_luts( + args.tokenizer_path, args.tokenizer_meta_path, args.vocab_size, device, + validate_meta=args.tokenizer_meta_validate, + ) + log0(f"tokenizer: kind={tokenizer_metadata.get('tokenizer_kind', 'unknown')} vocab={tokenizer_metadata.get('vocab_size', '?')}") + if tokenizer_metadata.get('tokenizer_kind') == 'sentencepiece': + sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) + if int(sp.vocab_size()) != args.vocab_size: + raise ValueError( + f"VOCAB_SIZE={args.vocab_size} does not match tokenizer vocab_size={int(sp.vocab_size())}" + ) + dataset_dir = Path(args.data_path).resolve() + actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) + effective_eval_seq_len = args.eval_seq_len if args.eval_seq_len > 0 else args.train_seq_len + val_seq_len = max(args.train_seq_len, effective_eval_seq_len) + val_tokens = load_validation_tokens(args.val_files, val_seq_len) + log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") + log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") + base_model = GPT( + vocab_size=args.vocab_size, + num_layers=args.num_layers, + model_dim=args.model_dim, + num_heads=args.num_heads, + num_kv_heads=args.num_kv_heads, + mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, + tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, + rope_base=args.rope_base, + qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, + bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, + ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, + ve_dim=args.ve_dim, + ve_layers=args.ve_layers, + neg_slope=args.negative_slope, + ).to(device).bfloat16() + # Banks stay FP32 (like CastedLinear weights), cast to BF16 in forward + base_model.qo_bank.data = base_model.qo_bank.data.float() + base_model.kv_bank.data = base_model.kv_bank.data.float() + base_model.mlp_up_bank.data = base_model.mlp_up_bank.data.float() + base_model.mlp_down_bank.data = base_model.mlp_down_bank.data.float() + for module in base_model.modules(): + if isinstance(module, CastedLinear): + module.float() + restore_low_dim_params_to_fp32(base_model) + # No DDP -- Parallel Muon handles bank grad communication via reduce-scatter, + # and non-bank grads are manually all-reduced before Adam steps. + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + model = compiled_model + + # Optimizer split: + # - 4 parameter banks -> Muon (batched Newton-Schulz) + # - token embedding -> Adam + # - scalars/control tensors -> Adam + # - bigram proj, VE proj -> Adam (small matrix params not worth banking) + matrix_params = [ + base_model.qo_bank, base_model.kv_bank, + base_model.mlp_up_bank, base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for name, p in block_named_params + if p.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + scalar_params.append(base_model.smear.gate) + if base_model.bigram is not None: + scalar_params.append(base_model.bigram.scale) + token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr + tok_params = [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}] + if base_model.bigram is not None: + tok_params.append({"params": [base_model.bigram.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.bigram.proj is not None: + scalar_params.append(base_model.bigram.proj.weight) + if base_model.ve_shared is not None: + tok_params.append({"params": [base_model.ve_shared.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.ve_shared.proj is not None: + scalar_params.append(base_model.ve_shared.proj.weight) + scalar_params.append(base_model.ve_shared.scale) + for s in base_model.ve_layer_scales: + scalar_params.append(s) + optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + optimizer_muon = Muon( + matrix_params, + lr=args.matrix_lr, + momentum=args.muon_momentum, + backend_steps=args.muon_backend_steps, + weight_decay=args.muon_wd, + ) + for group in optimizer_muon.param_groups: + group["base_lr"] = args.matrix_lr + optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + # Non-bank params that need manual all-reduce (replicated across GPUs) + replicated_params = list(optimizer_tok.param_groups[0]["params"]) + for pg in optimizer_tok.param_groups[1:]: + replicated_params.extend(pg["params"]) + replicated_params.extend(scalar_params) + + optimizer_head = None + if base_model.lm_head is not None: + optimizer_head = torch.optim.Adam( + [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + replicated_params.append(base_model.lm_head.weight) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if optimizer_head is not None: + optimizers.append(optimizer_head) + log0(f"model_params:{sum(p.numel() for p in base_model.parameters())}") + xsa_layers = [i for i, b in enumerate(base_model.blocks) if b.attn.use_xsa] + log0(f"XSA:last_{args.xsa_last_n} active_layers:{xsa_layers}") + log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") + log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") + log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") + log0( + f"tie_embeddings:{args.tie_embeddings} embed_lr:{token_lr} " + f"head_lr:{args.head_lr if base_model.lm_head is not None else 0.0} " + f"matrix_lr:{args.matrix_lr} scalar_lr:{args.scalar_lr}" + ) + log0( + f"train_batch_tokens:{args.train_batch_tokens} train_seq_len:{args.train_seq_len} " + f"iterations:{args.iterations} warmup_steps:{args.warmup_steps} " + f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" + ) + log0(f"seed:{args.seed}") + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + def zero_grad_all() -> None: + for opt in optimizers: + opt.zero_grad(set_to_none=True) + max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None + if args.use_gptq and max_wallclock_ms is not None: + max_wallclock_ms -= args.gptq_reserve_ms + log0(f"gptq:reserving {args.gptq_reserve_ms:.0f}ms from training budget, effective={max_wallclock_ms:.0f}ms") + def lr_mul(step: int, elapsed_ms: float) -> float: + if args.warmdown_iters <= 0: + return 1.0 + if max_wallclock_ms is None: + warmdown_start = max(args.iterations - args.warmdown_iters, 0) + return max((args.iterations - step) / max(args.warmdown_iters, 1), 0.0) if warmdown_start <= step < args.iterations else 1.0 + step_ms = elapsed_ms / max(step, 1) + warmdown_ms = args.warmdown_iters * step_ms + remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) + return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 + if args.warmup_steps > 0: + initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} + initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] + model.train() + for warmup_step in range(args.warmup_steps): + zero_grad_all() + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + warmup_loss = model(x, y) + (warmup_loss * grad_scale).backward() + # All-reduce all grads for warmup (simple, not optimized) + if distributed: + for p in base_model.parameters(): + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + for opt in optimizers: + opt.step() + zero_grad_all() + if args.warmup_steps <= 20 or (warmup_step + 1) % 10 == 0 or warmup_step + 1 == args.warmup_steps: + log0(f"warmup_step:{warmup_step + 1}/{args.warmup_steps}") + base_model.load_state_dict(initial_model_state, strict=True) + for opt, state in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + zero_grad_all() + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + swa_state: dict[str, Tensor] | None = None + swa_count = 0 + ema_state = {name: t.detach().float().clone() for name, t in base_model.state_dict().items()} + ema_decay = 0.997 + training_time_ms = 0.0 + stop_after_step: int | None = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) + should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1000.0 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + log0( + f"step:{step}/{args.iterations} val_loss:{val_loss:.4f} val_bpb:{val_bpb:.4f} " + f"train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms / max(step, 1):.2f}ms" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < args.iterations: + log0( + f"stopping_early: wallclock_cap train_time:{training_time_ms:.0f}ms " + f"step:{step}/{args.iterations}" + ) + break + elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + scale = lr_mul(step, elapsed_ms) + zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y) + train_loss += loss.detach() + (loss * grad_scale).backward() + train_loss /= grad_accum_steps + frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 + muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum + for group in optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * scale + if args.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) + # === 3-phase overlapped optimizer step === + # Phase 1: Launch async reduce-scatter for banks (biggest first) + optimizer_muon.launch_reduce_scatters() + # Phase 2: All-reduce non-bank grads + step Adam (while bank RS is in-flight) + if distributed: + for p in replicated_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + optimizer_tok.step() + optimizer_scalar.step() + if optimizer_head is not None: + optimizer_head.step() + # Phase 3: Wait for RS, local NS5, all-gather (banks processed last) + optimizer_muon.step() + zero_grad_all() + # EMA update + with torch.no_grad(): + for name, t in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_(t.detach().float(), alpha=1.0 - ema_decay) + step += 1 + approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + if args.swa_enabled and scale < 0.2 and step % args.swa_every == 0: + if swa_state is None: + swa_state = {name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()} + swa_count = 1 + log0(f"swa:start step:{step}") + else: + for name, t in base_model.state_dict().items(): + swa_state[name] += t.detach().cpu() + swa_count += 1 + should_log_train = ( + args.train_log_every > 0 + and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) + ) + if should_log_train: + log0( + f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " + f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" + ) + reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + if distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log0( + f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " + f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" + ) + # Apply EMA weights + log0("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = {name: t.to(dtype=current_state[name].dtype) for name, t in ema_state.items()} + base_model.load_state_dict(avg_state, strict=True) + torch.cuda.synchronize() + t_diag = time.perf_counter() + diag_val_loss, diag_val_bpb = eval_val( + args, compiled_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"DIAGNOSTIC post_ema val_loss:{diag_val_loss:.4f} val_bpb:{diag_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_diag):.0f}ms" + ) + export_sd = base_model.state_dict() + if master_process: + torch.save(export_sd, "final_model.pt") + model_bytes = os.path.getsize("final_model.pt") + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model: {model_bytes} bytes") + log0(f"Code size: {code_bytes} bytes") + # Unbank 3D tensors into individual 2D tensors for quantization + sd_cpu = {k: v.detach().cpu() for k, v in export_sd.items()} + unbanked_sd = _unbank_state_dict(sd_cpu, args.num_layers) + # GPTQ calibration: collect Hessians from training data + gptq_hessians = None + if args.use_gptq: + t_gptq = time.perf_counter() + log0(f"gptq:calibrating with {args.gptq_calib_samples} batches (training data)...") + calib_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + gptq_hessians = gptq_collect_hessians( + base_model, calib_loader, device, num_batches=args.gptq_calib_samples, + batch_tokens=args.train_batch_tokens, seq_len=args.train_seq_len, + grad_accum_steps=grad_accum_steps) + del calib_loader + gptq_elapsed = time.perf_counter() - t_gptq + log0(f"gptq:calibrated {len(gptq_hessians)} layers in {gptq_elapsed:.1f}s") + torch.cuda.empty_cache() + quant_result, quant_meta = mixed_quantize_int6(unbanked_sd, {"mlp", "attn"}, clip_range=args.quant_clip_range, hessians=gptq_hessians) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = lzma.compress(quant_raw, preset=6) + if master_process: + with open("final_model.int6.ptz", "wb") as f: + f.write(quant_blob) + quant_file_bytes = len(quant_blob) + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model int6+lzma: {quant_file_bytes} bytes") + log0(f"Total submission size int6+lzma: {quant_file_bytes + code_bytes} bytes") + if distributed: + dist.barrier() + with open("final_model.int6.ptz", "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(lzma.decompress(quant_blob_disk)), + map_location="cpu", + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd) + # Re-bank the dequantized tensors + deq_state = _rebank_state_dict(deq_unbanked, args.num_layers, sd_cpu) + eval_model = GPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + neg_slope=args.negative_slope, + ).to(device).bfloat16() + eval_model.qo_bank.data = eval_model.qo_bank.data.float() + eval_model.kv_bank.data = eval_model.kv_bank.data.float() + eval_model.mlp_up_bank.data = eval_model.mlp_up_bank.data.float() + eval_model.mlp_down_bank.data = eval_model.mlp_down_bank.data.float() + for m in eval_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(eval_model) + eval_model.load_state_dict(deq_state, strict=True) + compiled_eval = torch.compile(eval_model, dynamic=False, fullgraph=True) + torch.cuda.synchronize() + t_qeval = time.perf_counter() + q_val_loss, q_val_bpb = eval_val( + args, compiled_eval, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + eval_seq_len=effective_eval_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" + ) + log0(f"final_int6_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + sw_seq_len = effective_eval_seq_len + if args.eval_stride > 0 and args.eval_stride < sw_seq_len: + torch.cuda.synchronize() + t_slide = time.perf_counter() + sw_val_loss, sw_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window val_loss:{sw_val_loss:.4f} val_bpb:{sw_val_bpb:.4f} " + f"stride:{args.eval_stride} eval_time:{1000.0 * (time.perf_counter() - t_slide):.0f}ms" + ) + log0(f"final_int6_sliding_window_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + if args.eval_stride != 64 and 64 < sw_seq_len: + torch.cuda.synchronize() + t_slide64 = time.perf_counter() + sw64_val_loss, sw64_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=64, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window_s64 val_loss:{sw64_val_loss:.4f} val_bpb:{sw64_val_bpb:.4f} " + f"stride:64 eval_time:{1000.0 * (time.perf_counter() - t_slide64):.0f}ms" + ) + log0(f"final_int6_sliding_window_s64_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + # Legal score-first TTT (PR #461 recipe) + if args.ttt_enabled: + torch.cuda.synchronize() + t_ttt = time.perf_counter() + ttt_loss, ttt_bpb = eval_val_sliding_ttt( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, log0=log0, + ) + torch.cuda.synchronize() + log0(f"legal_ttt val_loss:{ttt_loss:.4f} val_bpb:{ttt_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt):.0f}ms") + log0(f"legal_ttt_exact val_loss:{ttt_loss:.8f} val_bpb:{ttt_bpb:.8f}") + if distributed: + dist.destroy_process_group() +if __name__ == "__main__": + main() + +==================================================================================================== +Running Python 3.12.3 (main, Mar 3 2026, 12:15:18) [GCC 13.3.0] +Running PyTorch 2.9.1+cu128 +Tue Mar 31 15:24:47 2026 ++-----------------------------------------------------------------------------------------+ +| NVIDIA-SMI 580.126.09 Driver Version: 580.126.09 CUDA Version: 13.0 | ++-----------------------------------------+------------------------+----------------------+ +| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | +| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | +| | | MIG M. | +|=========================================+========================+======================| +| 0 NVIDIA H100 80GB HBM3 On | 00000000:0A:00.0 Off | 0 | +| N/A 38C P0 119W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 1 NVIDIA H100 80GB HBM3 On | 00000000:18:00.0 Off | 0 | +| N/A 32C P0 125W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 2 NVIDIA H100 80GB HBM3 On | 00000000:3F:00.0 Off | 0 | +| N/A 31C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 3 NVIDIA H100 80GB HBM3 On | 00000000:48:00.0 Off | 0 | +| N/A 40C P0 127W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 4 NVIDIA H100 80GB HBM3 On | 00000000:87:00.0 Off | 0 | +| N/A 37C P0 122W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 5 NVIDIA H100 80GB HBM3 On | 00000000:90:00.0 Off | 0 | +| N/A 30C P0 114W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 6 NVIDIA H100 80GB HBM3 On | 00000000:BE:00.0 Off | 0 | +| N/A 30C P0 116W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 7 NVIDIA H100 80GB HBM3 On | 00000000:C7:00.0 Off | 0 | +| N/A 38C P0 123W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ + ++-----------------------------------------------------------------------------------------+ +| Processes: | +| GPU GI CI PID Type Process name GPU Memory | +| ID ID Usage | +|=========================================================================================| +| 0 N/A N/A 63786 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 1 N/A N/A 63787 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 2 N/A N/A 63788 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 3 N/A N/A 63789 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 4 N/A N/A 63790 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 5 N/A N/A 63791 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 6 N/A N/A 63792 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 7 N/A N/A 63793 C ...ameter-golf/.venv/bin/python3 1512MiB | ++-----------------------------------------------------------------------------------------+ + +==================================================================================================== +tokenizer: kind=tokenmonster vocab=998 +train_loader:dataset:fineweb10B_scylla train_shards:194 +val_loader:shards pattern=./data/datasets/fineweb10B_scylla/fineweb_val_*.bin tokens:62363648 +model_params:27022172 +XSA:last_11 active_layers:[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] +world_size:8 grad_accum_steps:1 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_mode:gqa num_heads:8 num_kv_heads:4 +tie_embeddings:True embed_lr:0.035 head_lr:0.0 matrix_lr:0.025 scalar_lr:0.025 +train_batch_tokens:786432 train_seq_len:2048 iterations:9000 warmup_steps:20 max_wallclock_seconds:600.000 +seed:2025 +gptq:reserving 9000ms from training budget, effective=591000ms +warmup_step:1/20 +warmup_step:2/20 +warmup_step:3/20 +warmup_step:4/20 +warmup_step:5/20 +warmup_step:6/20 +warmup_step:7/20 +warmup_step:8/20 +warmup_step:9/20 +warmup_step:10/20 +warmup_step:11/20 +warmup_step:12/20 +warmup_step:13/20 +warmup_step:14/20 +warmup_step:15/20 +warmup_step:16/20 +warmup_step:17/20 +warmup_step:18/20 +warmup_step:19/20 +warmup_step:20/20 +step:0/9000 val_loss:6.8931 val_bpb:3.3925 train_time:0ms step_avg:0.01ms +step:1/9000 train_loss:6.8925 train_time:122ms step_avg:121.74ms +step:2/9000 train_loss:9.2176 train_time:155ms step_avg:77.61ms +step:3/9000 train_loss:8.9342 train_time:242ms step_avg:80.53ms +step:4/9000 train_loss:8.4847 train_time:327ms step_avg:81.64ms +step:5/9000 train_loss:8.0678 train_time:411ms step_avg:82.16ms +step:6/9000 train_loss:7.5826 train_time:496ms step_avg:82.60ms +step:7/9000 train_loss:7.0192 train_time:581ms step_avg:83.07ms +step:8/9000 train_loss:6.7110 train_time:667ms step_avg:83.40ms +step:9/9000 train_loss:6.6151 train_time:752ms step_avg:83.58ms +step:10/9000 train_loss:6.4055 train_time:837ms step_avg:83.68ms +step:500/9000 train_loss:2.3213 train_time:43601ms step_avg:87.20ms +step:1000/9000 train_loss:2.2074 train_time:87405ms step_avg:87.41ms +step:1500/9000 train_loss:2.1863 train_time:131253ms step_avg:87.50ms +step:2000/9000 train_loss:2.1492 train_time:175187ms step_avg:87.59ms +step:2500/9000 train_loss:2.2069 train_time:219134ms step_avg:87.65ms +step:3000/9000 train_loss:2.1050 train_time:263091ms step_avg:87.70ms +step:3500/9000 train_loss:2.0852 train_time:307079ms step_avg:87.74ms +step:4000/9000 train_loss:2.0364 train_time:351032ms step_avg:87.76ms +step:4000/9000 val_loss:2.0777 val_bpb:1.0226 train_time:351087ms step_avg:87.77ms +step:4500/9000 train_loss:2.0086 train_time:395014ms step_avg:87.78ms +step:5000/9000 train_loss:1.9688 train_time:438965ms step_avg:87.79ms +step:5500/9000 train_loss:1.9353 train_time:482910ms step_avg:87.80ms +step:6000/9000 train_loss:1.9898 train_time:526848ms step_avg:87.81ms +swa:start step:6050 +step:6500/9000 train_loss:1.9760 train_time:571439ms step_avg:87.91ms +step:6719/9000 val_loss:1.9633 val_bpb:0.9663 train_time:591097ms step_avg:87.97ms +stopping_early: wallclock_cap train_time:591097ms step:6719/9000 +peak memory allocated: 23031 MiB reserved: 23324 MiB +ema:applying EMA weights +DIAGNOSTIC post_ema val_loss:1.9617 val_bpb:0.9655 eval_time:2111ms +Serialized model: 106201974 bytes +Code size: 101929 bytes +gptq:calibrating with 64 batches (training data)... +gptq:calibrated 66 layers in 6.7s +Serialized model int6+lzma: 15874004 bytes +Total submission size int6+lzma: 15975933 bytes +final_int6_roundtrip val_loss:1.9677 val_bpb:0.9684 eval_time:5337ms +final_int6_roundtrip_exact val_loss:1.96774016 val_bpb:0.96844526 +final_int6_sliding_window val_loss:1.9280 val_bpb:0.9489 stride:64 eval_time:76754ms +final_int6_sliding_window_exact val_loss:1.92801508 val_bpb:0.94888552 +final_int8_zlib_roundtrip_exact val_loss:1.92801508 val_bpb:0.94888552 diff --git a/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed42.log b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed42.log new file mode 100644 index 0000000000..a3609416eb --- /dev/null +++ b/records/track_10min_16mb/2026-03-31_Scylla_FullGPTQ_XSA11_FA3_0.9485/train_seed42.log @@ -0,0 +1,2308 @@ +from __future__ import annotations +import copy +import glob +import io +import lzma +import math +import os +import random +import subprocess +import sys +import time +import uuid +from pathlib import Path +import numpy as np +import sentencepiece as spm +import torch +import torch.distributed as dist +import torch.nn.functional as F +from torch import Tensor, nn +from torch.nn.parallel import DistributedDataParallel as DDP +from flash_attn_interface import flash_attn_func as flash_attn_3_func +class Hyperparameters: + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp1024") + train_files = os.path.join(data_path, "fineweb_train_*.bin") + val_files = os.path.join(data_path, "fineweb_val_*.bin") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_1024_bpe.model") + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + seed = int(os.environ.get("SEED", 1337)) + val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 3500)) + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786_432)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 1.5)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 1024)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 3.0)) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.035)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.025)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.025)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.99)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + swa_enabled = bool(int(os.environ.get("SWA_ENABLED", "1"))) + swa_every = int(os.environ.get("SWA_EVERY", 50)) + muon_wd = float(os.environ.get("MUON_WD", 0.04)) + adam_wd = float(os.environ.get("ADAM_WD", 0.04)) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 2048)) + bigram_dim = int(os.environ.get("BIGRAM_DIM", 128)) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 4)) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + ve_enabled = bool(int(os.environ.get("VE_ENABLED", "1"))) + ve_dim = int(os.environ.get("VE_DIM", 128)) + ve_layers = os.environ.get("VE_LAYERS", "9,10") + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "0"))) + ttt_lr = float(os.environ.get("TTT_LR", 0.002)) + ttt_epochs = int(os.environ.get("TTT_EPOCHS", 3)) + ttt_chunk_tokens = int(os.environ.get("TTT_CHUNK_TOKENS", 32768)) + ttt_freeze_blocks = int(os.environ.get("TTT_FREEZE_BLOCKS", 2)) + ttt_momentum = float(os.environ.get("TTT_MOMENTUM", 0.9)) + ttt_batch_seqs = int(os.environ.get("TTT_BATCH_SEQS", 32)) + ttt_grad_clip = float(os.environ.get("TTT_GRAD_CLIP", 1.0)) + negative_slope = float(os.environ.get("NEGATIVE_SLOPE", 0.5)) + use_gptq = bool(int(os.environ.get("USE_GPTQ", "0"))) + gptq_calib_samples = int(os.environ.get("GPTQ_CALIB_SAMPLES", "64")) + gptq_reserve_ms = float(os.environ.get("GPTQ_RESERVE_MS", "14000")) + quant_clip_range = int(os.environ.get("QUANT_CLIP_RANGE", 31)) + tokenizer_meta_path = os.environ.get("TOKENIZER_META_PATH", "") + tokenizer_meta_validate = bool(int(os.environ.get("TOKENIZER_META_VALIDATE", "0"))) + +# --- Batched Newton-Schulz orthogonalization --- + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 5, eps: float = 1e-7) -> Tensor: + """Batched Newton-Schulz orthogonalization. G: (B,M,N) or (M,N).""" + a, b, c = (3.4445, -4.7750, 2.0315) + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + for _ in range(steps): + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + +# --- Parallel Muon optimizer --- + +class Muon(torch.optim.Optimizer): + """Parallel Muon: post-backward reduce-scatter -> local NS5 -> all-gather. + + No DDP for bank params. After backward, this optimizer: + 1. Launches async reduce-scatter for all banks (biggest first) + 2. Returns control so Adam can step on small params while RS is in-flight + 3. Waits for each RS, runs local NS5 on the shard, launches async all-gather + 4. Each all-gather overlaps with next bank's NS5 + """ + def __init__(self, params, lr: float, momentum: float, backend_steps: int, + nesterov: bool = True, weight_decay: float = 0.0): + super().__init__( + params, + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, + nesterov=nesterov, weight_decay=weight_decay), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + 'p': p, + 'B': B, + 'padded_grad': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard_mom': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'full_update': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'scale': max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + # Sort by size descending -- launch biggest reduce-scatters first + self._bank_meta.sort(key=lambda m: -m['p'].numel()) + self._built = True + + def launch_reduce_scatters(self): + """Phase 1: launch async reduce-scatter for all banks. Call right after backward.""" + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m['p'] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m['padded_grad'] + pg[:m['B']].copy_(p.grad.bfloat16()) + if pg.shape[0] > m['B']: + pg[m['B']:].zero_() + fut = dist.reduce_scatter_tensor(m['shard'], pg, op=dist.ReduceOp.AVG, async_op=True) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + """Phase 3: wait for RS, local NS5, all-gather. Call AFTER Adam steps.""" + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + + if not self._built: + self._build() + + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + + prev_ag_handle = None + prev_m = None + + sharded = self._distributed and hasattr(self, '_rs_futures') + + for i, m in enumerate(self._bank_meta): + p = m['p'] + if p.grad is None: + continue + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if sharded and self._rs_futures[i] is not None: + self._rs_futures[i].wait() + g = m['shard'] + buf = m['shard_mom'] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m['full_update'], update, async_op=True) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m['scale']) + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if hasattr(self, '_rs_futures'): + del self._rs_futures + + return loss + +# --- Tokenizer evaluation helpers --- + +def build_sentencepiece_luts( + sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device +) -> tuple[Tensor, Tensor, Tensor]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) + +TOKENIZER_META_FORMAT_VERSION = 1 +TOKENIZER_META_SUFFIX = ".meta.npz" + + +def _derive_tokenizer_meta_path(tokenizer_path: str) -> Path: + tokenizer = Path(tokenizer_path) + if tokenizer.suffix == ".model": + return tokenizer.with_suffix(TOKENIZER_META_SUFFIX) + return tokenizer.with_name(tokenizer.name + TOKENIZER_META_SUFFIX) + + +def build_sentencepiece_luts_np( + sp: spm.SentencePieceProcessor, vocab_size: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return base_bytes_np, has_leading_space_np, is_boundary_token_np + + +def load_tokenizer_meta_luts_np( + meta_path: Path, vocab_size: int +) -> tuple[np.ndarray, np.ndarray, np.ndarray, dict[str, object]]: + def _scalar(value): + arr = np.asarray(value) + if arr.ndim == 0: + return arr.item() + first = arr.reshape(-1)[0] + return first.item() if hasattr(first, "item") else first + + with np.load(meta_path, allow_pickle=False) as data: + format_version = int(_scalar(data["format_version"])) + if format_version != TOKENIZER_META_FORMAT_VERSION: + raise ValueError( + f"Unsupported tokenizer meta format_version={format_version} " + f"expected={TOKENIZER_META_FORMAT_VERSION}" + ) + meta_vocab_size = int(_scalar(data["vocab_size"])) + tokenizer_kind = str(_scalar(data["tokenizer_kind"])) + source_model_name = str(_scalar(data["source_model_name"])) + base_bytes_np = np.asarray(data["base_bytes"], dtype=np.int16) + has_leading_space_np = np.asarray(data["has_leading_space"], dtype=np.bool_) + is_boundary_token_np = np.asarray(data["is_boundary_token"], dtype=np.bool_) + table_size = max(meta_vocab_size, vocab_size) + if base_bytes_np.shape[0] < table_size: + padded_base_bytes = np.zeros((table_size,), dtype=np.int16) + padded_has_leading_space = np.zeros((table_size,), dtype=np.bool_) + padded_is_boundary = np.ones((table_size,), dtype=np.bool_) + padded_base_bytes[: base_bytes_np.shape[0]] = base_bytes_np + padded_has_leading_space[: has_leading_space_np.shape[0]] = has_leading_space_np + padded_is_boundary[: is_boundary_token_np.shape[0]] = is_boundary_token_np + base_bytes_np = padded_base_bytes + has_leading_space_np = padded_has_leading_space + is_boundary_token_np = padded_is_boundary + metadata = { + "format_version": format_version, + "tokenizer_kind": tokenizer_kind, + "source_model_name": source_model_name, + "vocab_size": meta_vocab_size, + "meta_path": str(meta_path), + } + return base_bytes_np, has_leading_space_np, is_boundary_token_np, metadata + + +def load_tokenizer_luts( + tokenizer_path: str, + tokenizer_meta_path: str, + vocab_size: int, + device: torch.device, + *, + validate_meta: bool = False, +) -> tuple[tuple[Tensor, Tensor, Tensor], dict[str, object]]: + meta_path = ( + Path(tokenizer_meta_path) if tokenizer_meta_path + else _derive_tokenizer_meta_path(tokenizer_path) + ) + if meta_path.exists(): + base_bytes_np, has_leading_space_np, is_boundary_token_np, metadata = ( + load_tokenizer_meta_luts_np(meta_path, vocab_size) + ) + if validate_meta and str(tokenizer_path).endswith(".model"): + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + sp_luts = build_sentencepiece_luts_np(sp, vocab_size) + if not ( + np.array_equal(base_bytes_np, sp_luts[0]) + and np.array_equal(has_leading_space_np, sp_luts[1]) + and np.array_equal(is_boundary_token_np, sp_luts[2]) + ): + raise ValueError(f"Tokenizer metadata mismatch for {meta_path}") + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ), metadata + if not str(tokenizer_path).endswith(".model"): + raise FileNotFoundError(f"TOKENIZER_META_PATH does not exist: {meta_path}") + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + return build_sentencepiece_luts(sp, vocab_size, device), { + "tokenizer_kind": "sentencepiece", + "source_model_name": str(tokenizer_path), + "vocab_size": int(sp.vocab_size()), + "meta_path": None, + "fallback": True, + } + +def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = ((tokens.numel() - 1) // seq_len) * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] +def eval_val( + args: Hyperparameters, + model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + grad_accum_steps: int, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + seq_len = eval_seq_len or args.train_seq_len + local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + "VAL_BATCH_SIZE must provide at least one sequence per rank; " + f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " + f"GRAD_ACCUM_STEPS={grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + seq_start = (total_seqs * rank) // world_size + seq_end = (total_seqs * (rank + 1)) // world_size + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + model.eval() + with torch.inference_mode(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + batch_loss = model(x, y).detach() + batch_token_count = float(y.numel()) + val_loss_sum += batch_loss.to(torch.float64) * batch_token_count + val_token_count += batch_token_count + prev_ids = x.reshape(-1) + tgt_ids = y.reshape(-1) + token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + val_loss = val_loss_sum / val_token_count + bits_per_token = val_loss.item() / math.log(2.0) + tokens_per_byte = val_token_count.item() / val_byte_count.item() + model.train() + return float(val_loss.item()), float(bits_per_token * tokens_per_byte) + +# --- Quantization helpers --- + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,smear,dtg_gate,ve_layer_scales,ve_shared.scale,attn_gate,vr_lambda", + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_FP32_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "INT8_KEEP_FLOAT_FP32_NAME_PATTERNS", + ",".join(CONTROL_TENSOR_NAME_PATTERNS), + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_MAX_NUMEL = 65_536 +INT8_KEEP_FLOAT_STORE_DTYPE = torch.float16 +INT8_PER_ROW_SCALE_DTYPE = torch.float16 +INT8_CLIP_PERCENTILE = 99.99984 +INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 +def tensor_nbytes(t: Tensor) -> int: + return int(t.numel()) * int(t.element_size()) +def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: + if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): + return t.float().contiguous() + if t.dtype in {torch.float32, torch.bfloat16}: + passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") + return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() + return t +def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + clip_abs = ( + torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) + if t32.numel() + else torch.empty((t32.shape[0],), dtype=torch.float32) + ) + clipped = torch.maximum(torch.minimum(t32, clip_abs[:, None]), -clip_abs[:, None]) + scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) + q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() + return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() + clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 + scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) + q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() + return q, scale +def quantize_state_dict_int8(state_dict: dict[str, Tensor]): + quantized: dict[str, Tensor] = {} + scales: dict[str, Tensor] = {} + dtypes: dict[str, str] = {} + passthrough: dict[str, Tensor] = {} + passthrough_orig_dtypes: dict[str, str] = {} + qmeta: dict[str, dict[str, object]] = {} + stats = dict.fromkeys( + ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), + 0, + ) + for name, tensor in state_dict.items(): + t = tensor.detach().to("cpu").contiguous() + stats["param_count"] += int(t.numel()) + stats["num_tensors"] += 1 + stats["baseline_tensor_bytes"] += tensor_nbytes(t) + if not t.is_floating_point(): + stats["num_nonfloat_tensors"] += 1 + passthrough[name] = t + stats["int8_payload_bytes"] += tensor_nbytes(t) + continue + if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: + kept = keep_float_tensor(name, t, passthrough_orig_dtypes) + passthrough[name] = kept + stats["int8_payload_bytes"] += tensor_nbytes(kept) + continue + stats["num_float_tensors"] += 1 + q, s = quantize_float_tensor(t) + if s.ndim > 0: + qmeta[name] = {"scheme": "per_row", "axis": 0} + quantized[name] = q + scales[name] = s + dtypes[name] = str(t.dtype).removeprefix("torch.") + stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) + obj: dict[str, object] = { + "__quant_format__": "int8_clean_per_row_v1", + "quantized": quantized, + "scales": scales, + "dtypes": dtypes, + "passthrough": passthrough, + } + if qmeta: + obj["qmeta"] = qmeta + if passthrough_orig_dtypes: + obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes + return obj, stats +def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + qmeta = obj.get("qmeta", {}) + passthrough_orig_dtypes = obj.get("passthrough_orig_dtypes", {}) + for name, q in obj["quantized"].items(): + dtype = getattr(torch, obj["dtypes"][name]) + s = obj["scales"][name] + if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: + s = s.to(dtype=torch.float32) + out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() + else: + scale = float(s.item()) + out[name] = (q.float() * scale).to(dtype=dtype).contiguous() + for name, t in obj["passthrough"].items(): + out_t = t.detach().to("cpu").contiguous() + orig_dtype = passthrough_orig_dtypes.get(name) + if isinstance(orig_dtype, str): + out_t = out_t.to(dtype=getattr(torch, orig_dtype)).contiguous() + out[name] = out_t + return out + +# --- Data loading --- + +def load_data_shard(file: Path) -> Tensor: + header_bytes = 256 * np.dtype(" int: + key = str(file) + cached = _SHARD_NTOKENS_CACHE.get(key) + if cached is not None: + return cached + header = np.fromfile(file, dtype=" np.memmap: + key = str(file) + mm = _MMAP_CACHE.get(key) + if mm is not None: + return mm + n = _read_num_tokens(file) + mm = np.memmap(file, mode="r", dtype=" int: + if n <= 1: + return 1 + while True: + s = int(self._rng.integers(1, n)) + if math.gcd(s, n) == 1: + return s + def _reset_cursor(self, si: int, seq_len: int) -> None: + nt = int(self._num_tokens[si]) + max_phase = min(seq_len - 1, max(0, nt - seq_len - 1)) + phase = int(self._rng.integers(max_phase + 1)) if max_phase > 0 else 0 + bc = (nt - 1 - phase) // seq_len + self._cursor_phase[si] = phase + self._cursor_block_count[si] = bc + self._cursor_next[si] = 0 + self._cursor_start[si] = int(self._rng.integers(bc)) if bc > 1 else 0 + self._cursor_stride[si] = self._pick_coprime_stride(bc) + self._cursor_init[si] = True + def _ensure_cursor(self, si: int, seq_len: int) -> None: + if not self._cursor_init[si] or self._cursor_next[si] >= self._cursor_block_count[si]: + self._reset_cursor(si, seq_len) + def _take_from_shard(self, si: int, seq_len: int, count: int, out: list[tuple[int, int]]) -> None: + rem = count + while rem > 0: + self._ensure_cursor(si, seq_len) + bc = int(self._cursor_block_count[si]) + ni = int(self._cursor_next[si]) + take = min(rem, bc - ni) + phase = int(self._cursor_phase[si]) + start = int(self._cursor_start[si]) + stride = int(self._cursor_stride[si]) + for j in range(take): + bi = (start + (ni + j) * stride) % bc + out.append((si, phase + bi * seq_len)) + self._cursor_next[si] = ni + take + rem -= take + def _init_pipeline(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> None: + local_tokens = global_tokens // (self.world_size * grad_accum_steps) + num_seqs = local_tokens // seq_len + global_num_seqs = num_seqs * self.world_size + self._cfg = (local_tokens, seq_len, num_seqs, global_num_seqs) + bbc = (self._num_tokens - 1) // seq_len + eligible = bbc > 0 + self._eligible_shards = np.nonzero(eligible)[0].astype(np.int64) + self._base_block_counts = bbc[self._eligible_shards].astype(np.int64) + def _sample_global_windows(self) -> list[tuple[int, int]]: + assert self._cfg is not None and self._eligible_shards is not None + _, seq_len, _, gns = self._cfg + ec = int(self._eligible_shards.size) + progress = min(self._batches_built / 1800.0, 1.0) + remaining = np.empty(ec, dtype=np.float64) + for i, si in enumerate(self._eligible_shards.tolist()): + if self._cursor_init[si]: + r = int(self._cursor_block_count[si]) - int(self._cursor_next[si]) + remaining[i] = float(max(r, 1)) + else: + remaining[i] = float(self._base_block_counts[i]) + alpha = 0.90 - 0.40 * progress + weights = np.power(remaining, alpha) + ws = float(weights.sum()) + if not np.isfinite(ws) or ws <= 0.0: + weights = np.ones(ec, dtype=np.float64) + ws = float(weights.sum()) + probs = weights / ws + low = min(max(8, self.world_size), ec, gns) + high = min(max(32, self.world_size * 8), ec, gns) + mix = max(1, min(int(round(low + progress * (high - low))), ec, gns)) + cp = self._rng.choice(ec, size=mix, replace=False, p=probs) + cs = self._eligible_shards[cp] + cpr = probs[cp].copy() + cpr /= cpr.sum() + counts = np.ones(mix, dtype=np.int64) + extra = gns - mix + if extra > 0: + counts += self._rng.multinomial(extra, cpr).astype(np.int64) + perm = self._rng.permutation(mix) + cs, counts = cs[perm], counts[perm] + buckets: list[list[tuple[int, int]]] = [] + for si, cnt in zip(cs.tolist(), counts.tolist()): + b: list[tuple[int, int]] = [] + self._take_from_shard(int(si), seq_len, int(cnt), b) + if b: + if len(b) > 1: + bp = self._rng.permutation(len(b)) + b = [b[int(k)] for k in bp.tolist()] + buckets.append(b) + windows: list[tuple[int, int]] = [] + active = [i for i, bk in enumerate(buckets) if bk] + while active: + order = self._rng.permutation(len(active)) + new_active: list[int] = [] + for oi in order.tolist(): + bi = active[oi] + if buckets[bi]: + windows.append(buckets[bi].pop()) + if buckets[bi]: + new_active.append(bi) + active = new_active + return windows + def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: + if self._cfg is None: + self._init_pipeline(global_tokens, seq_len, grad_accum_steps) + _, _, num_seqs, gns = self._cfg + gw = self._sample_global_windows() + local_w = gw[self.rank::self.world_size] + x = torch.empty((num_seqs, seq_len), dtype=torch.int64) + y = torch.empty((num_seqs, seq_len), dtype=torch.int64) + for slot, (si, pos) in enumerate(local_w): + mm = _get_shard_memmap(self.files[si]) + window = torch.as_tensor(np.array(mm[pos:pos + seq_len + 1], dtype=np.int64)) + x[slot] = window[:-1] + y[slot] = window[1:] + self._batches_built += 1 + return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) + +# --- Transformer modules --- + +class RMSNorm(nn.Module): + def __init__(self, eps: float | None = None): + super().__init__() + self.eps = eps + def forward(self, x: Tensor) -> Tensor: + return F.rms_norm(x, (x.size(-1),), eps=self.eps) +class CastedLinear(nn.Linear): + def forward(self, x: Tensor) -> Tensor: + w = self.weight.to(x.dtype) + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) +def restore_low_dim_params_to_fp32(module: nn.Module) -> None: + with torch.no_grad(): + for name, param in module.named_parameters(): + if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: + param.data = param.data.float() +class Rotary(nn.Module): + def __init__(self, dim: int, base: float = 10000.0, train_seq_len: int = 1024, rope_dims: int = 0): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / (base ** (torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached: Tensor | None = None + self._sin_cached: Tensor | None = None + def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached != seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * (scale ** (rd / (rd - 2))) + inv_freq = 1.0 / (new_base ** (torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd)) + else: + inv_freq = self.inv_freq.to(device) + t = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor, rope_dims: int = 0) -> Tensor: + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + +class CausalSelfAttention(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + rope_base: float, + qk_gain_init: float, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + # No CastedLinear -- weights come from banks + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 # set by GPT.__init__ for partial RoPE + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False # set by GPT.__init__ for deep layers only + def _xsa_efficient(self, y: Tensor, v: Tensor) -> Tensor: + """Efficient XSA: subtract self-value projection via GQA-aware reshape (no repeat_interleave). + y: [B, T, H, D], v: [B, T, Hkv, D]. H must be divisible by Hkv.""" + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) # [B, T, Hkv, group, D] + vn = F.normalize(v, dim=-1).unsqueeze(-2) # [B, T, Hkv, 1, D] -- broadcast ready + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, v_embed: Tensor | None = None) -> Tensor: + if getattr(self, '_save_gptq', False): + self._gptq_qkv_in = x.detach() + bsz, seqlen, dim = x.shape + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, v_w.to(x.dtype)) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + y = flash_attn_3_func(q, k, v, causal=True) + if self.use_xsa: + y = self._xsa_efficient(y, v) + y = y.reshape(bsz, seqlen, dim) + if getattr(self, '_save_gptq', False): + self._gptq_o_in = y.detach() + return F.linear(y, out_w.to(x.dtype)) + +class SmearGate(nn.Module): + def __init__(self, dim: int): + super().__init__() + self.gate = nn.Parameter(torch.zeros(dim, dtype=torch.float32)) + def forward(self, x: Tensor) -> Tensor: + g = torch.sigmoid(self.gate.to(dtype=x.dtype))[None, None, :] + x_prev = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) + return (1 - g) * x + g * x_prev + +class BigramHashEmbedding(nn.Module): + def __init__(self, bigram_vocab_size: int, bigram_dim: int, model_dim: int): + super().__init__() + self.bigram_vocab_size = bigram_vocab_size + self.embed = nn.Embedding(bigram_vocab_size, bigram_dim) + nn.init.zeros_(self.embed.weight) + self.proj = CastedLinear(bigram_dim, model_dim, bias=False) if bigram_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.05, dtype=torch.float32)) + def bigram_hash(self, tokens: Tensor) -> Tensor: + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., 0] = mod + out[..., 1:] = torch.bitwise_xor(36313 * t[..., 1:], 27191 * t[..., :-1]) % mod + return out.long() + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(self.bigram_hash(token_ids)) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class ValueEmbedding(nn.Module): + """Reinject token identity into attention values at specific layers. + Each table maps vocab tokens to a low-dim embedding, projected to model_dim.""" + def __init__(self, vocab_size: int, ve_dim: int, model_dim: int): + super().__init__() + self.embed = nn.Embedding(vocab_size, ve_dim) + nn.init.normal_(self.embed.weight, std=0.01) + self.proj = CastedLinear(ve_dim, model_dim, bias=False) if ve_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.1, dtype=torch.float32)) + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(token_ids) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class MLP(nn.Module): + def __init__(self, dim: int, mlp_mult: int, neg_slope: float = 0.5): + super().__init__() + self.neg_slope = neg_slope + # No CastedLinear -- weights come from banks + def forward(self, x: Tensor, up_w: Tensor, down_w: Tensor) -> Tensor: + if getattr(self, '_save_gptq', False): + self._gptq_up_in = x.detach() + x = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=self.neg_slope) + x2 = x.square() + if getattr(self, '_save_gptq', False): + self._gptq_down_in = x2.detach() + return F.linear(x2, down_w.to(x.dtype)) + +class Block(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + rope_base: float, + qk_gain_init: float, + layer_idx: int = 0, + ln_scale: bool = False, + neg_slope: float = 0.5, + ): + super().__init__() + self.layer_idx = layer_idx + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = MLP(dim, mlp_mult, neg_slope=neg_slope) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + def forward(self, x: Tensor, x0: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, up_w: Tensor, down_w: Tensor, v_embed: Tensor | None = None) -> Tensor: + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, q_w, k_w, v_w, out_w, v_embed=v_embed) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + mlp_out = self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + return x_out + mlp_out + +class GPT(nn.Module): + def __init__( + self, + vocab_size: int, + num_layers: int, + model_dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + tie_embeddings: bool, + tied_embed_init_std: float, + logit_softcap: float, + rope_base: float, + qk_gain_init: float, + bigram_vocab_size: int = 0, + bigram_dim: int = 128, + xsa_last_n: int = 0, + rope_dims: int = 0, + ln_scale: bool = False, + ve_enabled: bool = False, + ve_dim: int = 128, + ve_layers: str = "9,10", + neg_slope: float = 0.5, + ): + super().__init__() + self._ve_target_dim = num_kv_heads * (model_dim // num_heads) # kv_dim for value projection + if logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") + self.tie_embeddings = tie_embeddings + self.tied_embed_init_std = tied_embed_init_std + self.logit_softcap = logit_softcap + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + # Parameter banks: contiguous 3D tensors for batched optimizer + head_dim = model_dim // num_heads + kv_dim = num_kv_heads * head_dim + mlp_dim = int(mlp_mult * model_dim) + self.num_layers = num_layers + self.qo_bank = nn.Parameter(torch.empty(2 * num_layers, model_dim, model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * num_layers, kv_dim, model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(num_layers, mlp_dim, model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(num_layers, model_dim, mlp_dim)) + self.blocks = nn.ModuleList( + [ + Block( + model_dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + layer_idx=i, + ln_scale=ln_scale, + neg_slope=neg_slope, + ) + for i in range(num_layers) + ] + ) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + kv_dim_ve = self._ve_target_dim + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim_ve) + self.ve_layer_scales = nn.ParameterList( + [nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices] + ) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.value_embeds = nn.ModuleList() # keep empty for compat + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + if self.lm_head is not None: + self.lm_head._zero_init = True + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + self._init_weights() + def _init_weights(self) -> None: + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + # Init banks: orthogonal, with proj layers scaled down and out/down zero-init + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) # Q + nn.init.zeros_(self.qo_bank.data[n + i]) # Out (zero init) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) # K + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) # V + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) # MLP up + nn.init.zeros_(self.mlp_down_bank.data[i]) # MLP down (zero init) + # Scale proj layers (out_proj and mlp_down are "proj" layers) + self.qo_bank.data[n + i].mul_(proj_scale) + self.mlp_down_bank.data[i].mul_(proj_scale) + # Init remaining nn.Linear modules (bigram proj, lm_head) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif module.weight.ndim == 2 and module.weight.shape[0] >= 64 and module.weight.shape[1] >= 64: + nn.init.orthogonal_(module.weight, gain=1.0) + def _get_ve(self, layer_idx: int, input_ids: Tensor, ve_cache: dict | None = None) -> Tensor | None: + """Get value embedding for a specific layer using shared table + per-layer scale.""" + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if ve_cache is not None and 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_base = ve_cache['ve'] if ve_cache is not None else self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_base * self.ve_layer_scales[ve_idx].to(dtype=ve_base.dtype) + def forward(self, input_ids: Tensor, target_ids: Tensor) -> Tensor: + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + if self.tie_embeddings: + logits_proj = F.linear(x_flat, self.tok_emb.weight) + else: + if self.lm_head is None: + raise RuntimeError("lm_head is required when tie_embeddings=False") + logits_proj = self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy(logits.float(), targets, reduction="mean") + def forward_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve) + x = self.final_norm(x) + if self.tie_embeddings: + logits_proj = F.linear(x, self.tok_emb.weight) + else: + logits_proj = self.lm_head(x) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + +# --- Sliding window evaluation --- + +def eval_val_sliding( + args: Hyperparameters, + base_model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + stride: int, + batch_seqs: int = 32, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + """Sliding window evaluation: each token scored with maximum context.""" + seq_len = eval_seq_len or args.train_seq_len + total_tokens = val_tokens.numel() - 1 + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= 1] + total_windows = len(window_starts) + my_s = (total_windows * rank) // world_size + my_e = (total_windows * (rank + 1)) // world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + base_model.eval() + compiled_logits = torch.compile(base_model.forward_logits, dynamic=False, fullgraph=True) + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk[:-1] + y_batch[i, :wlen] = chunk[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = compiled_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), + reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt = y_batch[i, s:wlen] + prev = x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + val_loss = (loss_sum / token_count).item() + bits_per_token = val_loss / math.log(2.0) + tokens_per_byte = token_count.item() / byte_count.item() + base_model.train() + return val_loss, bits_per_token * tokens_per_byte + + +def eval_val_sliding_ttt( + args: Hyperparameters, base_model: nn.Module, rank: int, world_size: int, + device: torch.device, val_tokens: Tensor, base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, is_boundary_token_lut: Tensor, + stride: int, batch_seqs: int = 32, log0=print, +) -> tuple[float, float]: + """Legal score-first TTT (PR #461 recipe): score each chunk with sliding windows, + then train on it. Every token scored BEFORE any update that could use it.""" + seq_len = args.train_seq_len + total_tokens = val_tokens.numel() - 1 + ttt_chunk = args.ttt_chunk_tokens + + # Pre-compute all window starts + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= stride or ws == 0] + + # Assign each window to a chunk based on the first token it scores + num_chunks = (total_tokens + ttt_chunk - 1) // ttt_chunk + chunk_windows: list[list[int]] = [[] for _ in range(num_chunks)] + for ws in window_starts: + end = min(ws + seq_len, total_tokens) + wlen = end - ws + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_start = ws + s + ci = min(scored_start // ttt_chunk, num_chunks - 1) + chunk_windows[ci].append(ws) + + log0(f"ttt_sliding:start chunks={num_chunks} chunk_tokens={ttt_chunk} " + f"total_windows={len(window_starts)} stride={stride} " + f"ttt_lr={args.ttt_lr} ttt_epochs={args.ttt_epochs} " + f"freeze_blocks={args.ttt_freeze_blocks}") + + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + + # Freeze first N blocks + frozen_block_ids = set(range(min(args.ttt_freeze_blocks, len(base_model.blocks)))) + ttt_params = [] + for name, p in base_model.named_parameters(): + freeze = False + for bi in frozen_block_ids: + if f"blocks.{bi}." in name: + freeze = True + break + if freeze: + p.requires_grad_(False) + else: + p.requires_grad_(True) + ttt_params.append(p) + + log0(f"ttt_sliding:params unfrozen={sum(p.numel() for p in ttt_params)} " + f"frozen={sum(p.numel() for p in base_model.parameters() if not p.requires_grad)}") + + optimizer = torch.optim.SGD(ttt_params, lr=args.ttt_lr, momentum=args.ttt_momentum) + t0 = time.perf_counter() + + for ci in range(num_chunks): + windows = chunk_windows[ci] + if not windows: + continue + chunk_start = ci * ttt_chunk + chunk_end = min((ci + 1) * ttt_chunk, total_tokens) + + # --- Phase 1: SCORE this chunk's windows (inference_mode) --- + my_s = (len(windows) * rank) // world_size + my_e = (len(windows) * (rank + 1)) // world_size + my_windows = windows[my_s:my_e] + + base_model.eval() + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk_tok = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk_tok[:-1] + y_batch[i, :wlen] = chunk_tok[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = base_model.forward_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt, prev = y_batch[i, s:wlen], x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + + # --- Phase 2: TRAIN on this chunk (already scored = legal) --- + is_last_chunk = (ci == num_chunks - 1) + if not is_last_chunk and args.ttt_epochs > 0: + base_model.train() + chunk_seqs = (chunk_end - chunk_start) // seq_len + if chunk_seqs > 0: + cos_lr = args.ttt_lr * 0.5 * (1.0 + math.cos(math.pi * ci / max(num_chunks - 1, 1))) + for pg in optimizer.param_groups: + pg['lr'] = cos_lr + my_seq_s = (chunk_seqs * rank) // world_size + my_seq_e = (chunk_seqs * (rank + 1)) // world_size + my_chunk_seqs = my_seq_e - my_seq_s + for _ep in range(args.ttt_epochs): + for bs in range(0, my_chunk_seqs, args.ttt_batch_seqs): + be = min(bs + args.ttt_batch_seqs, my_chunk_seqs) + actual_bs = my_seq_s + bs + start_tok = chunk_start + actual_bs * seq_len + end_tok = chunk_start + (my_seq_s + be) * seq_len + 1 + if end_tok > val_tokens.numel(): + continue + local = val_tokens[start_tok:end_tok].to(device=device, dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + optimizer.zero_grad(set_to_none=True) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + loss = base_model(x, y) + loss.backward() + if world_size > 1: + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + torch.nn.utils.clip_grad_norm_(ttt_params, args.ttt_grad_clip) + optimizer.step() + + if rank == 0 and (ci % 10 == 0 or ci == num_chunks - 1): + elapsed = time.perf_counter() - t0 + rl = loss_sum.item() / max(token_count.item(), 1) + rbpb = rl / math.log(2.0) * (token_count.item() / max(byte_count.item(), 1)) if token_count.item() > 0 else 0.0 + log0(f" ttt_chunk [{ci+1}/{num_chunks}] bpb={rbpb:.6f} time={elapsed:.1f}s") + + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_count.item()) + + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + + log0(f"ttt_sliding:done val_loss={val_loss:.6f} val_bpb={val_bpb:.6f} " + f"elapsed={time.perf_counter() - t0:.1f}s") + return val_loss, val_bpb + + +# --- GPTQ-lite int6 quantization --- + +def _classify_param(name: str) -> str: + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or (".proj." in name and ".mlp." not in name): + return "attn" + return "other" +def quantize_int6_per_row(t: Tensor, clip_range: int = 31) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + best_q, best_s, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(t32.abs(), pct, dim=1) + else: + row_clip = t32.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + recon = q.float() * s.float()[:, None] + err = (t32 - recon).pow(2).mean().item() + if err < best_err: + best_q, best_s, best_err = q, s, err + return best_q, best_s + amax = t32.abs().max().item() + scale = torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + q = torch.clamp(torch.round(t32 / scale.float()), -clip_range, clip_range).to(torch.int8) + return q, scale + +def _unbank_state_dict(sd: dict[str, Tensor], num_layers: int) -> dict[str, Tensor]: + """Convert 3D bank tensors into individual 2D tensors with standard names.""" + out: dict[str, Tensor] = {} + n = num_layers + for name, tensor in sd.items(): + if name == "qo_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_q.weight"] = tensor[i] + out[f"blocks.{i}.attn.proj.weight"] = tensor[n + i] + elif name == "kv_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_k.weight"] = tensor[i] + out[f"blocks.{i}.attn.c_v.weight"] = tensor[n + i] + elif name == "mlp_up_bank": + for i in range(n): + out[f"blocks.{i}.mlp.fc.weight"] = tensor[i] + elif name == "mlp_down_bank": + for i in range(n): + out[f"blocks.{i}.mlp.proj.weight"] = tensor[i] + else: + out[name] = tensor + return out + +def _rebank_state_dict(sd: dict[str, Tensor], num_layers: int, template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + """Convert individual 2D tensors back into 3D bank tensors.""" + out: dict[str, Tensor] = {} + n = num_layers + # Reconstruct banks from individual weight keys + qo_slices = [None] * (2 * n) + kv_slices = [None] * (2 * n) + up_slices = [None] * n + down_slices = [None] * n + consumed = set() + for i in range(n): + qk = f"blocks.{i}.attn.c_q.weight" + if qk in sd: + qo_slices[i] = sd[qk] + consumed.add(qk) + ok = f"blocks.{i}.attn.proj.weight" + if ok in sd: + qo_slices[n + i] = sd[ok] + consumed.add(ok) + kk = f"blocks.{i}.attn.c_k.weight" + if kk in sd: + kv_slices[i] = sd[kk] + consumed.add(kk) + vk = f"blocks.{i}.attn.c_v.weight" + if vk in sd: + kv_slices[n + i] = sd[vk] + consumed.add(vk) + fk = f"blocks.{i}.mlp.fc.weight" + if fk in sd: + up_slices[i] = sd[fk] + consumed.add(fk) + dk = f"blocks.{i}.mlp.proj.weight" + if dk in sd: + down_slices[i] = sd[dk] + consumed.add(dk) + out["qo_bank"] = torch.stack(qo_slices).to(dtype=template_sd["qo_bank"].dtype) + out["kv_bank"] = torch.stack(kv_slices).to(dtype=template_sd["kv_bank"].dtype) + out["mlp_up_bank"] = torch.stack(up_slices).to(dtype=template_sd["mlp_up_bank"].dtype) + out["mlp_down_bank"] = torch.stack(down_slices).to(dtype=template_sd["mlp_down_bank"].dtype) + for name, tensor in sd.items(): + if name not in consumed: + out[name] = tensor + return out + +def mixed_quantize_int6(state_dict: dict[str, Tensor], int6_cats: set[str], clip_range: int = 31, + hessians: dict[str, Tensor] | None = None): + num_layers_total = max( + (int(k.split(".")[1]) for k in state_dict if k.startswith("blocks.")), + default=0, + ) + 1 + late_k_layers = set(range(num_layers_total - 2, num_layers_total)) + result: dict[str, Tensor] = {} + meta: dict[str, object] = {} + gptq_count, naive_count = 0, 0 + for name, tensor in state_dict.items(): + t = tensor.detach().cpu().contiguous() + cat = _classify_param(name) + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough" + continue + if any(p in name for p in CONTROL_TENSOR_NAME_PATTERNS): + result[name] = t.float() + meta[name] = "passthrough_ctrl" + continue + if cat in int6_cats and t.ndim >= 1: + H = hessians.get(name) if hessians else None + if H is not None and t.ndim == 2: + q, s = gptq_quantize_weight(t, H.cpu(), clip_range=clip_range) + gptq_count += 1 + else: + q, s = quantize_int6_per_row(t, clip_range=clip_range) + naive_count += 1 + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int6"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + if hessians: + print(f"gptq_quantize: {gptq_count} GPTQ layers, {naive_count} naive layers", flush=True) + return result, meta +def dequantize_mixed_int6(result: dict[str, Tensor], meta: dict[str, object], + template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + for name, orig in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if info in ("passthrough", "passthrough_ctrl", "passthrough_fp16"): + t = result[name] + if t.dtype == torch.float16 and orig_dtype in (torch.float32, torch.bfloat16): + t = t.to(orig_dtype) + out[name] = t + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out + +# --- Full Hessian GPTQ --- + +def gptq_quantize_weight(W: Tensor, H: Tensor, clip_range: int = 31, + block_size: int = 128, percdamp: float = 0.01) -> tuple[Tensor, Tensor]: + """GPTQ with Cholesky error compensation and actorder (Frantar et al., ICLR 2023).""" + W_orig = W.float().clone() + rows, cols = W_orig.shape + H = H.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = percdamp * H.diag().mean() + H.diagonal().add_(damp) + perm = torch.argsort(H.diag(), descending=True) + invperm = torch.argsort(perm) + W_perm = W_orig[:, perm].clone() + W_perm[:, dead[perm]] = 0 + H = H[perm][:, perm] + try: + Hinv = torch.cholesky_inverse(torch.linalg.cholesky(H)) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + except torch.linalg.LinAlgError: + return quantize_int6_per_row(W_orig, clip_range) + best_q, best_scale, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(W_orig.abs(), pct, dim=1) + else: + row_clip = W_orig.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + sf = s.float() + Q = torch.zeros(rows, cols, dtype=torch.int8) + W_work = W_perm.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + W_block = W_work[:, i1:i2].clone() + Hinv_block = Hinv[i1:i2, i1:i2] + Err = torch.zeros(rows, i2 - i1) + for j in range(i2 - i1): + w_col = W_block[:, j] + d = Hinv_block[j, j] + q_col = torch.clamp(torch.round(w_col / sf), -clip_range, clip_range) + Q[:, i1 + j] = q_col.to(torch.int8) + err = (w_col - q_col.float() * sf) / d + Err[:, j] = err + W_block[:, j:] -= err.unsqueeze(1) * Hinv_block[j, j:].unsqueeze(0) + if i2 < cols: + W_work[:, i2:] -= Err @ Hinv[i1:i2, i2:] + recon = Q.float() * sf[:, None] + mse = (W_perm - recon).pow(2).mean().item() + if mse < best_err: + best_q, best_scale, best_err = Q, s, mse + best_q = best_q[:, invperm] + return best_q, best_scale + +def _init_hessians(nl: int, dim: int, mlp_dim: int, device: torch.device) -> dict[str, Tensor]: + h: dict[str, Tensor] = {} + for i in range(nl): + for k in ['c_q', 'c_k', 'c_v']: + h[f'blocks.{i}.attn.{k}.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.attn.proj.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.mlp.fc.weight'] = torch.zeros(dim, dim, dtype=torch.float32, device=device) + h[f'blocks.{i}.mlp.proj.weight'] = torch.zeros(mlp_dim, mlp_dim, dtype=torch.float32, device=device) + return h + +def _accum_hessians(hessians: dict[str, Tensor], blocks: nn.ModuleList, dim: int, mlp_dim: int) -> None: + for i, block in enumerate(blocks): + qkv_in = block.attn._gptq_qkv_in.float().reshape(-1, dim) + h_qkv = qkv_in.t() @ qkv_in + hessians[f'blocks.{i}.attn.c_q.weight'] += h_qkv + hessians[f'blocks.{i}.attn.c_k.weight'] += h_qkv + hessians[f'blocks.{i}.attn.c_v.weight'] += h_qkv + o_in = block.attn._gptq_o_in.float().reshape(-1, dim) + hessians[f'blocks.{i}.attn.proj.weight'] += o_in.t() @ o_in + up_in = block.mlp._gptq_up_in.float().reshape(-1, dim) + hessians[f'blocks.{i}.mlp.fc.weight'] += up_in.t() @ up_in + down_in = block.mlp._gptq_down_in.float().reshape(-1, mlp_dim) + hessians[f'blocks.{i}.mlp.proj.weight'] += down_in.t() @ down_in + +def _finalize_hessians(hessians: dict[str, Tensor], num_batches: int) -> None: + for name in hessians: + hessians[name] = hessians[name].cpu() / num_batches + damp = 0.01 * torch.diag(hessians[name]).mean().clamp_min(1e-6) + hessians[name] += damp * torch.eye(hessians[name].shape[0]) + +def gptq_collect_hessians(base_model: nn.Module, train_loader, device: torch.device, + num_batches: int, batch_tokens: int, seq_len: int, + grad_accum_steps: int) -> dict[str, Tensor]: + """Collect Hessians H = X^T X from training data.""" + nl = base_model.num_layers + dim = base_model.tok_emb.weight.shape[1] + mlp_dim = base_model.mlp_up_bank.shape[1] + hessians = _init_hessians(nl, dim, mlp_dim, device) + for block in base_model.blocks: + block.attn._save_gptq = True + block.mlp._save_gptq = True + base_model.eval() + with torch.inference_mode(), torch.autocast(device_type='cuda', dtype=torch.bfloat16): + for _ in range(num_batches): + x, y = train_loader.next_batch(batch_tokens, seq_len, grad_accum_steps) + base_model(x, y) + _accum_hessians(hessians, base_model.blocks, dim, mlp_dim) + for block in base_model.blocks: + block.attn._save_gptq = False + block.mlp._save_gptq = False + _finalize_hessians(hessians, num_batches) + base_model.train() + return hessians + +# --- Training --- + +def main() -> None: + code = Path(__file__).read_text(encoding="utf-8") + args = Hyperparameters() + # zeropower_via_newtonschulz5 runs eagerly with bmm -- do NOT compile + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError(f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral") + grad_accum_steps = 8 // world_size + grad_scale = 1.0 / grad_accum_steps + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + master_process = rank == 0 + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + logfile = None + if master_process: + os.makedirs("logs", exist_ok=True) + logfile = f"logs/{args.run_id}.txt" + print(logfile) + def log0(msg: str, console: bool = True) -> None: + if not master_process: + return + if console: + print(msg) + if logfile is not None: + with open(logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + log0(code, console=False) + log0("=" * 100, console=False) + log0(f"Running Python {sys.version}", console=False) + log0(f"Running PyTorch {torch.__version__}", console=False) + log0( + subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=False).stdout, + console=False, + ) + log0("=" * 100, console=False) + random.seed(args.seed) + np.random.seed(args.seed) + torch.manual_seed(args.seed) + torch.cuda.manual_seed_all(args.seed) + (base_bytes_lut, has_leading_space_lut, is_boundary_token_lut), tokenizer_metadata = load_tokenizer_luts( + args.tokenizer_path, args.tokenizer_meta_path, args.vocab_size, device, + validate_meta=args.tokenizer_meta_validate, + ) + log0(f"tokenizer: kind={tokenizer_metadata.get('tokenizer_kind', 'unknown')} vocab={tokenizer_metadata.get('vocab_size', '?')}") + if tokenizer_metadata.get('tokenizer_kind') == 'sentencepiece': + sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) + if int(sp.vocab_size()) != args.vocab_size: + raise ValueError( + f"VOCAB_SIZE={args.vocab_size} does not match tokenizer vocab_size={int(sp.vocab_size())}" + ) + dataset_dir = Path(args.data_path).resolve() + actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) + effective_eval_seq_len = args.eval_seq_len if args.eval_seq_len > 0 else args.train_seq_len + val_seq_len = max(args.train_seq_len, effective_eval_seq_len) + val_tokens = load_validation_tokens(args.val_files, val_seq_len) + log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") + log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") + base_model = GPT( + vocab_size=args.vocab_size, + num_layers=args.num_layers, + model_dim=args.model_dim, + num_heads=args.num_heads, + num_kv_heads=args.num_kv_heads, + mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, + tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, + rope_base=args.rope_base, + qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, + bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, + ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, + ve_dim=args.ve_dim, + ve_layers=args.ve_layers, + neg_slope=args.negative_slope, + ).to(device).bfloat16() + # Banks stay FP32 (like CastedLinear weights), cast to BF16 in forward + base_model.qo_bank.data = base_model.qo_bank.data.float() + base_model.kv_bank.data = base_model.kv_bank.data.float() + base_model.mlp_up_bank.data = base_model.mlp_up_bank.data.float() + base_model.mlp_down_bank.data = base_model.mlp_down_bank.data.float() + for module in base_model.modules(): + if isinstance(module, CastedLinear): + module.float() + restore_low_dim_params_to_fp32(base_model) + # No DDP -- Parallel Muon handles bank grad communication via reduce-scatter, + # and non-bank grads are manually all-reduced before Adam steps. + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + model = compiled_model + + # Optimizer split: + # - 4 parameter banks -> Muon (batched Newton-Schulz) + # - token embedding -> Adam + # - scalars/control tensors -> Adam + # - bigram proj, VE proj -> Adam (small matrix params not worth banking) + matrix_params = [ + base_model.qo_bank, base_model.kv_bank, + base_model.mlp_up_bank, base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for name, p in block_named_params + if p.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + scalar_params.append(base_model.smear.gate) + if base_model.bigram is not None: + scalar_params.append(base_model.bigram.scale) + token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr + tok_params = [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}] + if base_model.bigram is not None: + tok_params.append({"params": [base_model.bigram.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.bigram.proj is not None: + scalar_params.append(base_model.bigram.proj.weight) + if base_model.ve_shared is not None: + tok_params.append({"params": [base_model.ve_shared.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.ve_shared.proj is not None: + scalar_params.append(base_model.ve_shared.proj.weight) + scalar_params.append(base_model.ve_shared.scale) + for s in base_model.ve_layer_scales: + scalar_params.append(s) + optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + optimizer_muon = Muon( + matrix_params, + lr=args.matrix_lr, + momentum=args.muon_momentum, + backend_steps=args.muon_backend_steps, + weight_decay=args.muon_wd, + ) + for group in optimizer_muon.param_groups: + group["base_lr"] = args.matrix_lr + optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + # Non-bank params that need manual all-reduce (replicated across GPUs) + replicated_params = list(optimizer_tok.param_groups[0]["params"]) + for pg in optimizer_tok.param_groups[1:]: + replicated_params.extend(pg["params"]) + replicated_params.extend(scalar_params) + + optimizer_head = None + if base_model.lm_head is not None: + optimizer_head = torch.optim.Adam( + [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + replicated_params.append(base_model.lm_head.weight) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if optimizer_head is not None: + optimizers.append(optimizer_head) + log0(f"model_params:{sum(p.numel() for p in base_model.parameters())}") + xsa_layers = [i for i, b in enumerate(base_model.blocks) if b.attn.use_xsa] + log0(f"XSA:last_{args.xsa_last_n} active_layers:{xsa_layers}") + log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") + log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") + log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") + log0( + f"tie_embeddings:{args.tie_embeddings} embed_lr:{token_lr} " + f"head_lr:{args.head_lr if base_model.lm_head is not None else 0.0} " + f"matrix_lr:{args.matrix_lr} scalar_lr:{args.scalar_lr}" + ) + log0( + f"train_batch_tokens:{args.train_batch_tokens} train_seq_len:{args.train_seq_len} " + f"iterations:{args.iterations} warmup_steps:{args.warmup_steps} " + f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" + ) + log0(f"seed:{args.seed}") + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + def zero_grad_all() -> None: + for opt in optimizers: + opt.zero_grad(set_to_none=True) + max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None + if args.use_gptq and max_wallclock_ms is not None: + max_wallclock_ms -= args.gptq_reserve_ms + log0(f"gptq:reserving {args.gptq_reserve_ms:.0f}ms from training budget, effective={max_wallclock_ms:.0f}ms") + def lr_mul(step: int, elapsed_ms: float) -> float: + if args.warmdown_iters <= 0: + return 1.0 + if max_wallclock_ms is None: + warmdown_start = max(args.iterations - args.warmdown_iters, 0) + return max((args.iterations - step) / max(args.warmdown_iters, 1), 0.0) if warmdown_start <= step < args.iterations else 1.0 + step_ms = elapsed_ms / max(step, 1) + warmdown_ms = args.warmdown_iters * step_ms + remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) + return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 + if args.warmup_steps > 0: + initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} + initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] + model.train() + for warmup_step in range(args.warmup_steps): + zero_grad_all() + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + warmup_loss = model(x, y) + (warmup_loss * grad_scale).backward() + # All-reduce all grads for warmup (simple, not optimized) + if distributed: + for p in base_model.parameters(): + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + for opt in optimizers: + opt.step() + zero_grad_all() + if args.warmup_steps <= 20 or (warmup_step + 1) % 10 == 0 or warmup_step + 1 == args.warmup_steps: + log0(f"warmup_step:{warmup_step + 1}/{args.warmup_steps}") + base_model.load_state_dict(initial_model_state, strict=True) + for opt, state in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + zero_grad_all() + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + swa_state: dict[str, Tensor] | None = None + swa_count = 0 + ema_state = {name: t.detach().float().clone() for name, t in base_model.state_dict().items()} + ema_decay = 0.997 + training_time_ms = 0.0 + stop_after_step: int | None = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) + should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1000.0 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + log0( + f"step:{step}/{args.iterations} val_loss:{val_loss:.4f} val_bpb:{val_bpb:.4f} " + f"train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms / max(step, 1):.2f}ms" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < args.iterations: + log0( + f"stopping_early: wallclock_cap train_time:{training_time_ms:.0f}ms " + f"step:{step}/{args.iterations}" + ) + break + elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + scale = lr_mul(step, elapsed_ms) + zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y) + train_loss += loss.detach() + (loss * grad_scale).backward() + train_loss /= grad_accum_steps + frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 + muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum + for group in optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * scale + if args.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) + # === 3-phase overlapped optimizer step === + # Phase 1: Launch async reduce-scatter for banks (biggest first) + optimizer_muon.launch_reduce_scatters() + # Phase 2: All-reduce non-bank grads + step Adam (while bank RS is in-flight) + if distributed: + for p in replicated_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + optimizer_tok.step() + optimizer_scalar.step() + if optimizer_head is not None: + optimizer_head.step() + # Phase 3: Wait for RS, local NS5, all-gather (banks processed last) + optimizer_muon.step() + zero_grad_all() + # EMA update + with torch.no_grad(): + for name, t in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_(t.detach().float(), alpha=1.0 - ema_decay) + step += 1 + approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + if args.swa_enabled and scale < 0.2 and step % args.swa_every == 0: + if swa_state is None: + swa_state = {name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()} + swa_count = 1 + log0(f"swa:start step:{step}") + else: + for name, t in base_model.state_dict().items(): + swa_state[name] += t.detach().cpu() + swa_count += 1 + should_log_train = ( + args.train_log_every > 0 + and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) + ) + if should_log_train: + log0( + f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " + f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" + ) + reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + if distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log0( + f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " + f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" + ) + # Apply EMA weights + log0("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = {name: t.to(dtype=current_state[name].dtype) for name, t in ema_state.items()} + base_model.load_state_dict(avg_state, strict=True) + torch.cuda.synchronize() + t_diag = time.perf_counter() + diag_val_loss, diag_val_bpb = eval_val( + args, compiled_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"DIAGNOSTIC post_ema val_loss:{diag_val_loss:.4f} val_bpb:{diag_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_diag):.0f}ms" + ) + export_sd = base_model.state_dict() + if master_process: + torch.save(export_sd, "final_model.pt") + model_bytes = os.path.getsize("final_model.pt") + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model: {model_bytes} bytes") + log0(f"Code size: {code_bytes} bytes") + # Unbank 3D tensors into individual 2D tensors for quantization + sd_cpu = {k: v.detach().cpu() for k, v in export_sd.items()} + unbanked_sd = _unbank_state_dict(sd_cpu, args.num_layers) + # GPTQ calibration: collect Hessians from training data + gptq_hessians = None + if args.use_gptq: + t_gptq = time.perf_counter() + log0(f"gptq:calibrating with {args.gptq_calib_samples} batches (training data)...") + calib_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + gptq_hessians = gptq_collect_hessians( + base_model, calib_loader, device, num_batches=args.gptq_calib_samples, + batch_tokens=args.train_batch_tokens, seq_len=args.train_seq_len, + grad_accum_steps=grad_accum_steps) + del calib_loader + gptq_elapsed = time.perf_counter() - t_gptq + log0(f"gptq:calibrated {len(gptq_hessians)} layers in {gptq_elapsed:.1f}s") + torch.cuda.empty_cache() + quant_result, quant_meta = mixed_quantize_int6(unbanked_sd, {"mlp", "attn"}, clip_range=args.quant_clip_range, hessians=gptq_hessians) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = lzma.compress(quant_raw, preset=6) + if master_process: + with open("final_model.int6.ptz", "wb") as f: + f.write(quant_blob) + quant_file_bytes = len(quant_blob) + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model int6+lzma: {quant_file_bytes} bytes") + log0(f"Total submission size int6+lzma: {quant_file_bytes + code_bytes} bytes") + if distributed: + dist.barrier() + with open("final_model.int6.ptz", "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(lzma.decompress(quant_blob_disk)), + map_location="cpu", + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd) + # Re-bank the dequantized tensors + deq_state = _rebank_state_dict(deq_unbanked, args.num_layers, sd_cpu) + eval_model = GPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + neg_slope=args.negative_slope, + ).to(device).bfloat16() + eval_model.qo_bank.data = eval_model.qo_bank.data.float() + eval_model.kv_bank.data = eval_model.kv_bank.data.float() + eval_model.mlp_up_bank.data = eval_model.mlp_up_bank.data.float() + eval_model.mlp_down_bank.data = eval_model.mlp_down_bank.data.float() + for m in eval_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(eval_model) + eval_model.load_state_dict(deq_state, strict=True) + compiled_eval = torch.compile(eval_model, dynamic=False, fullgraph=True) + torch.cuda.synchronize() + t_qeval = time.perf_counter() + q_val_loss, q_val_bpb = eval_val( + args, compiled_eval, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + eval_seq_len=effective_eval_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" + ) + log0(f"final_int6_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + sw_seq_len = effective_eval_seq_len + if args.eval_stride > 0 and args.eval_stride < sw_seq_len: + torch.cuda.synchronize() + t_slide = time.perf_counter() + sw_val_loss, sw_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window val_loss:{sw_val_loss:.4f} val_bpb:{sw_val_bpb:.4f} " + f"stride:{args.eval_stride} eval_time:{1000.0 * (time.perf_counter() - t_slide):.0f}ms" + ) + log0(f"final_int6_sliding_window_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + if args.eval_stride != 64 and 64 < sw_seq_len: + torch.cuda.synchronize() + t_slide64 = time.perf_counter() + sw64_val_loss, sw64_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=64, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window_s64 val_loss:{sw64_val_loss:.4f} val_bpb:{sw64_val_bpb:.4f} " + f"stride:64 eval_time:{1000.0 * (time.perf_counter() - t_slide64):.0f}ms" + ) + log0(f"final_int6_sliding_window_s64_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + # Legal score-first TTT (PR #461 recipe) + if args.ttt_enabled: + torch.cuda.synchronize() + t_ttt = time.perf_counter() + ttt_loss, ttt_bpb = eval_val_sliding_ttt( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, log0=log0, + ) + torch.cuda.synchronize() + log0(f"legal_ttt val_loss:{ttt_loss:.4f} val_bpb:{ttt_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt):.0f}ms") + log0(f"legal_ttt_exact val_loss:{ttt_loss:.8f} val_bpb:{ttt_bpb:.8f}") + if distributed: + dist.destroy_process_group() +if __name__ == "__main__": + main() + +==================================================================================================== +Running Python 3.12.3 (main, Mar 3 2026, 12:15:18) [GCC 13.3.0] +Running PyTorch 2.9.1+cu128 +Tue Mar 31 15:10:52 2026 ++-----------------------------------------------------------------------------------------+ +| NVIDIA-SMI 580.126.09 Driver Version: 580.126.09 CUDA Version: 13.0 | ++-----------------------------------------+------------------------+----------------------+ +| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | +| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | +| | | MIG M. | +|=========================================+========================+======================| +| 0 NVIDIA H100 80GB HBM3 On | 00000000:0A:00.0 Off | 0 | +| N/A 33C P0 118W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 1 NVIDIA H100 80GB HBM3 On | 00000000:18:00.0 Off | 0 | +| N/A 30C P0 123W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 2 NVIDIA H100 80GB HBM3 On | 00000000:3F:00.0 Off | 0 | +| N/A 28C P0 116W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 3 NVIDIA H100 80GB HBM3 On | 00000000:48:00.0 Off | 0 | +| N/A 35C P0 123W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 4 NVIDIA H100 80GB HBM3 On | 00000000:87:00.0 Off | 0 | +| N/A 33C P0 121W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 5 NVIDIA H100 80GB HBM3 On | 00000000:90:00.0 Off | 0 | +| N/A 28C P0 113W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 6 NVIDIA H100 80GB HBM3 On | 00000000:BE:00.0 Off | 0 | +| N/A 28C P0 115W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 7 NVIDIA H100 80GB HBM3 On | 00000000:C7:00.0 Off | 0 | +| N/A 34C P0 121W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ + ++-----------------------------------------------------------------------------------------+ +| Processes: | +| GPU GI CI PID Type Process name GPU Memory | +| ID ID Usage | +|=========================================================================================| +| 0 N/A N/A 62800 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 1 N/A N/A 62801 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 2 N/A N/A 62802 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 3 N/A N/A 62803 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 4 N/A N/A 62804 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 5 N/A N/A 62805 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 6 N/A N/A 62806 C ...ameter-golf/.venv/bin/python3 1512MiB | +| 7 N/A N/A 62807 C ...ameter-golf/.venv/bin/python3 1512MiB | ++-----------------------------------------------------------------------------------------+ + +==================================================================================================== +tokenizer: kind=tokenmonster vocab=998 +train_loader:dataset:fineweb10B_scylla train_shards:194 +val_loader:shards pattern=./data/datasets/fineweb10B_scylla/fineweb_val_*.bin tokens:62363648 +model_params:27022172 +XSA:last_11 active_layers:[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] +world_size:8 grad_accum_steps:1 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_mode:gqa num_heads:8 num_kv_heads:4 +tie_embeddings:True embed_lr:0.035 head_lr:0.0 matrix_lr:0.025 scalar_lr:0.025 +train_batch_tokens:786432 train_seq_len:2048 iterations:9000 warmup_steps:20 max_wallclock_seconds:600.000 +seed:42 +gptq:reserving 9000ms from training budget, effective=591000ms +warmup_step:1/20 +warmup_step:2/20 +warmup_step:3/20 +warmup_step:4/20 +warmup_step:5/20 +warmup_step:6/20 +warmup_step:7/20 +warmup_step:8/20 +warmup_step:9/20 +warmup_step:10/20 +warmup_step:11/20 +warmup_step:12/20 +warmup_step:13/20 +warmup_step:14/20 +warmup_step:15/20 +warmup_step:16/20 +warmup_step:17/20 +warmup_step:18/20 +warmup_step:19/20 +warmup_step:20/20 +step:0/9000 val_loss:6.8915 val_bpb:3.3918 train_time:0ms step_avg:0.01ms +step:1/9000 train_loss:6.8908 train_time:122ms step_avg:121.65ms +step:2/9000 train_loss:9.2384 train_time:153ms step_avg:76.72ms +step:3/9000 train_loss:8.9442 train_time:238ms step_avg:79.36ms +step:4/9000 train_loss:8.4911 train_time:324ms step_avg:81.04ms +step:5/9000 train_loss:8.0243 train_time:408ms step_avg:81.70ms +step:6/9000 train_loss:7.4500 train_time:494ms step_avg:82.27ms +step:7/9000 train_loss:6.9807 train_time:579ms step_avg:82.65ms +step:8/9000 train_loss:6.7101 train_time:664ms step_avg:83.05ms +step:9/9000 train_loss:6.5964 train_time:749ms step_avg:83.20ms +step:10/9000 train_loss:6.3878 train_time:834ms step_avg:83.39ms +step:500/9000 train_loss:2.3151 train_time:43554ms step_avg:87.11ms +step:1000/9000 train_loss:2.2086 train_time:87306ms step_avg:87.31ms +step:1500/9000 train_loss:2.1805 train_time:131209ms step_avg:87.47ms +step:2000/9000 train_loss:2.1469 train_time:175159ms step_avg:87.58ms +step:2500/9000 train_loss:2.2068 train_time:219040ms step_avg:87.62ms +step:3000/9000 train_loss:2.1005 train_time:262973ms step_avg:87.66ms +step:3500/9000 train_loss:2.0811 train_time:306855ms step_avg:87.67ms +step:4000/9000 train_loss:2.0310 train_time:350773ms step_avg:87.69ms +step:4000/9000 val_loss:2.0749 val_bpb:1.0212 train_time:350830ms step_avg:87.71ms +step:4500/9000 train_loss:2.0054 train_time:394663ms step_avg:87.70ms +step:5000/9000 train_loss:1.9694 train_time:438544ms step_avg:87.71ms +step:5500/9000 train_loss:1.9324 train_time:482435ms step_avg:87.72ms +step:6000/9000 train_loss:1.9929 train_time:526342ms step_avg:87.72ms +swa:start step:6050 +step:6500/9000 train_loss:1.9742 train_time:570896ms step_avg:87.83ms +step:6725/9000 val_loss:1.9610 val_bpb:0.9651 train_time:591069ms step_avg:87.89ms +stopping_early: wallclock_cap train_time:591069ms step:6725/9000 +peak memory allocated: 23031 MiB reserved: 23324 MiB +ema:applying EMA weights +DIAGNOSTIC post_ema val_loss:1.9595 val_bpb:0.9644 eval_time:2111ms +Serialized model: 106201974 bytes +Code size: 101929 bytes +gptq:calibrating with 64 batches (training data)... +gptq:calibrated 66 layers in 6.7s +Serialized model int6+lzma: 15869164 bytes +Total submission size int6+lzma: 15971093 bytes +final_int6_roundtrip val_loss:1.9655 val_bpb:0.9674 eval_time:5303ms +final_int6_roundtrip_exact val_loss:1.96553681 val_bpb:0.96736085 +final_int6_sliding_window val_loss:1.9255 val_bpb:0.9476 stride:64 eval_time:76779ms +final_int6_sliding_window_exact val_loss:1.92548833 val_bpb:0.94764197 +final_int8_zlib_roundtrip_exact val_loss:1.92548833 val_bpb:0.94764197