FINAL: val_bpb 1.1251 — artifact 15.90MB — within 16MB limit!

MLP 3.25x on 8xH100 SXM, 10 min:
- 5408 steps at 111ms/step
- Training val_bpb: 1.1455
- Int6 GPTQ roundtrip: 1.1485 (standard), 1.1251 (sliding s64)
- Artifact: 15.90MB (under 16MB limit!)
- Pruning: only 1 value (0.0%) — nearly fits without pruning

Leaderboard position: between openai#3 (1.1228) and openai#4 (1.1248)

Trinity innovation: wider MLP (3.25x vs SOTA 3x) from ternary
parameter budget analysis. All weights int6 GPTQ.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: 2 people

records/track_10min_16mb/2026-04-02_Trinity_Hybrid_Ternary_GPTQ_XSA/submission.json

@@ -4,8 +4,8 @@
   "name": "Trinity_Hybrid_MLP_XSA",
   "author": "gHashTag",
   "github_id": "deborahnelson8788726",
-  "val_bpb": 1.1279,
-  "val_bpb_note": "sliding window s64, MLP 3.5x, artifact 16.67MB (slightly over 16MB limit — MLP 3.25x expected to fit)",
+  "val_bpb": 1.1251,
+  "val_bpb_note": "sliding window s64, MLP 3.25x, artifact 15.90MB (within 16MB limit)",
   "description": "Trinity-inspired wider MLP (3.5x vs SOTA 3x) enabled by parameter budget analysis from ternary computing research. Built on PR #1019 stack (AR Self-Gen GPTQ, XSA-all, BigramHash, LeakyReLU², Partial RoPE, EMA/SWA). All weights quantized with int6 GPTQ.",
   "base": "2026-03-25_ValCalib_GPTQ_XSA_BigramHash3072",
   "architecture": "11L 512d 8h/4kv MLP3.25x int6-GPTQ",