Non-record: PR703 + shard-order curriculum + GPTQ cache-backout (1.1171)

[petergpt]

Summary

This PR adds a non-record 16MB submission built on the PR703-style quant/cache-backout branch with a score-ranked shard curriculum.

Submitted as non-record because:

• it is a single-seed package
• it does not clear the 0.005-nat record threshold required for leaderboard record promotion

Result

• final_int6_sliding_window_exact = 1.11709895
• final_int6_roundtrip_exact = 1.14068680
• post_ema = 1.1368
• step_stop = 6918
• step_avg = 86.75ms
• total submission size = 15,909,560
• bytes under cap = 90,440

What changed relative to the forked PR703 base

The base PR703 local package was 1.11748714 at 15,963,300 bytes. This submission keeps the same general object class and mainly changes:

1. score-ranked shard curriculum
2. tighter final int6+lzma packing

This is an incremental optimization package, not a new frontier architecture claim.

Included files

• README.md
• submission.json
• train.log
• train_gpt.py
• score_shards.py

Submission folder:

• records/track_non_record_16mb/2026-03-25_PR703_Curriculum_Carryover_1.1171

[MatoTeziTanka]

Community Review — Non-record: PR703 + shard-order curriculum + GPTQ cache-backout (1.1171)

BPB: (not parsed — see PR title) | Compliance: LOOKS CLEAN — score-first-per-chunk TTT (legal #1416/#1423 pattern)

What I found in the code (head SHA 8b46c09972b3, file records/track_non_record_16mb/2026-03-25_PR703_Curriculum_Carryover_1.1171/train_gpt.py):

The TTT path at line 1183 implements the score-first-per-chunk pattern: each chunk is scored under torch.no_grad() / inference_mode() before the base_model.train() + SGD adaptation runs on that same chunk, with an is_last_chunk guard so the final chunk gets no adaptation pass. This is the structural shape the legal frontier uses (PRs #1416 erichroepke, #1423 aryanbhosale).

Per Issue #402 and Issue #677, TTT is legal when each token is scored before the adapter updates on it, and that's what the code does here — chunk ci is scored under weights adapted only on chunks 0..ci-1. No prequant_ttt_adapt_adamw(val_tokens, ...) multi-epoch fine-tune, no scored-region SLOT, no target-in-key n-gram cache.

CPU smoke test (CT2038 proteus-engine, 2026-04-11): import OK in 0.06s, dim=512, layers=11, vocab=1024, code=111257 B, SMOKE_TEST_PASS

Verdict: LOOKS CLEAN.

Recommendation to @cocohearts @valerio-oai @0hq @yuzhougu-oai @notapplica: MERGE pending standard checks (3-seed validation, 16MB artifact cap, 10-min wallclock on 8×H100 SXM). The compliance picture matches the legal reference frontier and no flags were raised by the classification pass.

Auto-classification caveat: this review was drafted by the AST-based classifier against a template derived from manually-reviewed cluster PRs (#1420, #1450, #1487, #1541, #1529, #1533, #1518). If I've misread a subtlety in your eval path — e.g., multi-epoch TTT that I mistook for single-pass, or a target-in-key lookup I missed in a helper function — please flag it and I'll re-run the audit manually.

---

Reviewed by @MatoTeziTanka — The Agora. CPU smoke test (CT2038 proteus-engine, 2026-04-11): import OK in 0.06s, dim=512, layers=11, vocab=1024, code=111257 B, SMOKE_TEST_PASS. Classification via deterministic AST-based classify_prs.py (pattern bank derived from ~65 manually-reviewed PRs earlier in the 2026-04-11 sweep). This review was auto-drafted from a template and spot-checked before posting — if the template misread your code, please call it out so I can iterate the classifier.