Add DeepFloor non-record submission candidate

[KenMalloy]

Summary

This PR adds a new non-record submission candidate: 2026-04-03_DeepFloor.

DeepFloor is a compact recurrent multi-view language model that replaces a large flat stack with:

• repeated QKV+O recurrent blocks
• two cross-token modes: periodic floor attention or a fused recurrent accumulator
• explicit stability controls: contraction target, decay, TBPTT chunking, Jacobian penalty, periodic normalization, and optional HiPPO-style fused state cores
• a frozen record-folder submission snapshot so the submission entrypoint runs without importing mutable repo-root code

Why this is interesting

This is not a leaderboard-quality run yet. The value of the submission is that it packages a real, unusual architecture direction that is reproducible inside the challenge record format:

• depth recurrence / universal-transformer-style computation
• multi-view recurrence instead of standard layer stacking
• a self-contained submission path built for cold pods and remote validation

Included artifacts

• README.md
• RESULTS.md
• submission.json
• train_gpt.py
• deepfloor_snapshot.py
• freeze_submission_snapshot.py
• runner scripts for local, smallbox, and fullbox validation
• train log and candidate result json

Current checked-in evidence

Best real-enwik8 small-box matrix candidate:

• fused_d32_v2
• val_bpb = 7.9221
• artifact_bytes = 8448

Frozen submission preflight:

• bytes_total = 56221
• bytes_model_estimated = 8192
• bytes_code = 48029

Checked-in candidate submission:

• submission.json built from candidate_result_seed1337.json
• bytes_total = 56477
• val_bpb = 7.9221

Validation

• local submission preflight on the frozen record-folder snapshot
• remote GPU submission preflight on a reused small pod
• remote small-box suite on real enwik8
• remote 8x H100 fullbox suite, with synced smoke, matrix, launch_logs, and evolution artifacts under runs/fullbox/

Fullbox evidence

Best fullbox recipe-search result:

• frontier seed 2025
• val_bpb = 4.1101
• test_bpb = 4.0239
• artifact_mb = 0.1377
• mode: floor
• state core: scalar_decay
• recurrent dim: 96
• views: 8

This is stronger evidence that DeepFloor is a real direction worth reviewing, but it is still not presented as a record claim or as three repeated fixed-candidate contest runs.

Submission lane

This should be reviewed as a track_non_record_16mb research submission, not a record claim.

[MatoTeziTanka]

[RETRACTED 2026-04-11] — This IMPORT_FAIL was a false positive. Root cause: sibling module exists in same records/ folder; runner sys.path bug. Your code is not broken. See correction below: #1304 (comment)

---

Community Review — Add DeepFloor non-record submission candidate

Compliance: NEEDS AUTHOR ACTION — train_gpt.py fails to import on CT2038 (Python 3.10 / torch 2.10.0+cpu)

What I found: The CPU smoke test on CT2038 (proteus-engine, 128 GB RAM, Triton 3.6.0, flash_attn stub, cutlass_evt_fusion stub) failed at the import step with:

ModuleNotFoundError: No module named 'spectral_flood_walk_v0'

A few of the common patterns I've seen for this class of error in the 2026-04-11 sweep:

• PEP 701 f-string nesting — e.g. log(f" {cat}: {", ".join(...)}") is valid Python 3.12+ but invalid Python 3.10 because the inner ", " re-enters the outer double-quote context. One-character fix: ', ' instead of ", ". See PR Record: SP8192 + Improved Parallel Residuals + Muon 0.97 + LR 0.03 + Legal TTT — val_bpb 1.07785 (3-seed mean) #1541 / Record: SP8192 + Triple Recurrence + Banking + Fused MLP + Muon 0.97 — val_bpb 1.0778 (3-seed mean) #1523 for reference.
• Missing flash_attn variants — e.g. from flash_attn_interface import flash_attn_varlen_func when the wrapper script only stubs flash_attn_func. Not a PR defect on H100s, but the eval image / CPU preflight path needs a guarded import.
• Local compiled extension — e.g. import cutlass_evt_fusion from a records/*/cutlass_evt_fusion/ subfolder that isn't on the import path at smoke time. Usually an import-order issue inside the script.
• Actual syntax error — typo, missing bracket, etc.

Recommendation: Could you run python3 -c "import py_compile; py_compile.compile('train_gpt.py')" on your records-folder train_gpt.py under Python 3.10 specifically? The eval image is Python 3.10 per Issue #17 / the README, so any parse error on 3.10 blocks the submission at import time before any of the scored-eval logic runs.

Once the parse/import issue is fixed, I'll re-run the compliance audit through the normal pipeline. No other flags identified yet because the audit halts at the import step.

---

Reviewed by @MatoTeziTanka — The Agora. CPU smoke test (CT2038 proteus-engine, 2026-04-11): IMPORT_FAIL — ModuleNotFoundError: No module named 'spectral_flood_walk_v0'. Classification via classify_prs.py AST-based classifier; full compliance audit deferred until the import issue is resolved. Auto-drafted from a template and spot-checked before posting.

[MatoTeziTanka]

Retraction — this IMPORT_FAIL was a bug in my smoke runner

Sorry @KenMalloy, this one's on me. I re-audited the IMPORT_FAIL I posted above and it was a false positive — the fault is in how my CPU smoke runner set up sys.path, not in your code.

What happened:

The runner imported your ./train_gpt.py with only the script's folder implicitly on sys.path, so when your file did from spectral_flood_walk_v0 import ... it couldn't resolve the sibling spectral_flood_walk_v0.py that lives in the same ./ directory. The error I reported — ModuleNotFoundError: No module named 'spectral_flood_walk_v0' — looked like a missing file, but I re-checked the head SHA b3169d4 and spectral_flood_walk_v0.py is right there, committed to the PR, next to train_gpt.py.

Verified at head b3169d4:

spectral_flood_walk_v0.py   ← sibling module, exists
./train_gpt.py   ← imports it

On the real eval image (Python 3.10, records/*/ as the working dir), this import resolves correctly because the records folder ends up on sys.path via the standard cwd-driven import or via the eval harness's per-record entry point.

Your PR is not broken by this error. I'm retracting the IMPORT_FAIL classification. I'll re-queue the full compliance audit (BPB check, n-gram / TTT / SLOT flags, etc.) on the current head and post findings separately.

Again — sorry for the noise. These community reviews only work if I actually read what I'm reviewing, and I didn't in this case.

[KenMalloy]

This is a draft PR, and not my full bet. Please don't spend too much effort on it.

Keep up the good work.

[MatoTeziTanka]

Understood — noted for the backlog. The retraction of the IMPORT_FAIL still stands either way. When you're ready to promote it from draft, the compliance pass will pick it up in the next sweep.