[Non-record] Meta-Learned TTT + Error-Guided Adaptation Analysis (val_bpb=1.1645)#294
Closed
sseanliu wants to merge 12 commits intoopenai:mainfrom
Closed
[Non-record] Meta-Learned TTT + Error-Guided Adaptation Analysis (val_bpb=1.1645)#294sseanliu wants to merge 12 commits intoopenai:mainfrom
sseanliu wants to merge 12 commits intoopenai:mainfrom
Conversation
…x 4 recycled = 12 effective layers Architecture: 3 unique blocks at dim=768 (12 heads, 6 KV heads) recycled 4x each for 12 effective layers with per-iteration scale/mix params and U-Net skip connections. 13.2M unique params in ~12MB compressed (3.9MB headroom vs 16MB cap). 50% wider representations + 20% more effective depth vs SOTA's 10x512.
- Extract _apply_block helper for cleaner per-iteration logic - Remove _block_for_layer, use direct modular indexing - Reduce code from 1325 to 1321 lines
Three files: - program.md: Instructions for the AI agent (experiment loop, logging, directions) - prepare.py: Fixed utilities (data loading, evaluation, quantization, size checking) - train.py: Modifiable baseline (SOTA architecture, the only file the agent edits) Based on Karpey's autoresearch framework, adapted for parameter-golf constraints (16MB artifact limit, fixed FineWeb dataset, SentencePiece 1024 vocab).
…o MetaTTT submission
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Non-record research submission exploring test-time adaptation strategies for compressed language models at 16MB scale.
Key findings
Score
Methodology
Files
records/track_10min_16mb/2026-03-20_MetaTTT_v2/train_gpt.py— Training with Reptileeval_error_guided_ttt.py— Error-guided TTT evaluationrecords/track_10min_16mb/2026-03-20_MetaTTT_v2/README.md— Full analysisrecords/track_10min_16mb/2026-03-20_MetaTTT_v2/submission.json— MetadataSee README for detailed methodology, results, and theoretical context.