PR #180 SOTA: 10L Int5-MLP + BigramHash(10240) + SWA(0.4) + WD=0.04#1
Open
PR #180 SOTA: 10L Int5-MLP + BigramHash(10240) + SWA(0.4) + WD=0.04#1
Conversation
…0.04 Reproduce openai/parameter-golf PR openai#180 (val_bpb 1.14276, 3-seed mean). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace every other MLP (layers 0,2,4,6,8) with BigLU — an MLP where the hidden state is gated by a per-layer bigram embedding (vocab=2048, dim=hidden, expansion scale=1). Reduce mlp_mult 3.0→1.5 (hidden 1536→768) so total MLP params stay identical to PR openai#180 (15.73M). - Muon for up/down weights; AdamW for bigram embed tables (like main bigram) - bigram.embed excluded from matrix_params to avoid Muon Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add BIGLU_MULT env var (default 1.0): biglu_hidden = biglu_mult * dim = 512 independent of mlp_mult=3.0 (hidden=1536) — per-layer params identical - Switch BigLU activation from relu² to F.silu - run.sh/local.sh: drop MLP_MULT override, add BIGLU_MULT=1.0 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
BigLU bigram embed tables use token_lr * biglu_bigram_lr_mult for faster adaptation of the per-layer bigram gates. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
GRAD_ACCUM_STEPS env var overrides the default 8//world_size, enabling single-GPU runs that match run.sh's effective batch (786432 tokens/step). local.sh starts at GRAD_ACCUM_STEPS=64 (12 seqs×1024 micro-batch); increment to 128/192/384/768 if OOM. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…cal.sh eval_val was incorrectly dividing val batch by grad_accum_steps (a training concept irrelevant to eval), causing ValueError at high GRAD_ACCUM_STEPS. local.sh: GRAD_ACCUM_STEPS=128 (6 seqs × 1024 micro-batch, fits in 7.6GB). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
FA3 (flash_attn_interface, SM90/H100) → FA2 (flash_attn, GQA only when num_kv_heads==num_heads) → PyTorch SDPA fallback with enable_gqa. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Reproduces openai/parameter-golf PR openai#180.
Summary
train_gpt.py(val_bpb 1.14276, 3-seed mean)run.shfor 8xH100 full reproduction (SEED=1337 bash run.shfor specific seed)local.shfor 1-GPU local test (reduced batch/wallclock)PR openai#180 Techniques
Metrics
🤖 Generated with Claude Code