Add depth recurrence + SwiGLU submission (Apple M3 8GB)#8
Closed
iranzithierry wants to merge 2 commits intoopenai:mainfrom
Closed
Add depth recurrence + SwiGLU submission (Apple M3 8GB)#8iranzithierry wants to merge 2 commits intoopenai:mainfrom
iranzithierry wants to merge 2 commits intoopenai:mainfrom
Conversation
Add a non-record leaderboard submission exploring depth recurrence and SwiGLU MLPs trained on an Apple M3 (8GB). Includes README with architecture/hyperparameter notes, submission.json metadata, and a full training script (train_gpt_mlx.py) implementing: 4 unique transformer blocks reused 3× (12 effective layers), SwiGLU MLP, wider 640-dim model, per-recurrence gates, U-Net skips, gradient clipping, split optimizers (Muon + Adam), token streaming, and int8+zlib quantization/roundtrip. Notes limitations from hardware and guidance to run on larger hardware for competitive results.
Collaborator
|
You need a train.log and a val_bpb for a non-record submission! |
gb250e
referenced
this pull request
in gb250e/parameter-golf
Mar 21, 2026
dhruvjatkar
pushed a commit
to dhruvjatkar/parameter-golf
that referenced
this pull request
Mar 25, 2026
PR openai#672 maxes TTT at 30 epochs (590s/600s eval budget), so all future improvements must be orthogonal to TTT. This update: - Sets 1.0781 BPB (PR openai#672) as the new target to beat - Reorders Top 8 directions: XSA-all confirmed at #1, Full GPTQ #2, SwiGLU #3, Muon-VS #4, aggressive quant #5, MASA openai#6, depth recurrence openai#7 with int6 risk warning, AdEMAMix openai#8 - Deprioritizes TTT-related directions already exploited by PR openai#672 - Collapses ~1000 lines of stale Round 0-3.9 session logs into a concise historical summary - Removes resolved blockers (flash_attn, SSH hangs, local runtime) - Adds fresh Round 1 section with 5 submitted experiments Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
dhruvjatkar
pushed a commit
to dhruvjatkar/parameter-golf
that referenced
this pull request
Mar 25, 2026
PR openai#672 maxes TTT at 30 epochs (590s/600s eval budget), so all future improvements must be orthogonal to TTT. This update: - Sets 1.0781 BPB (PR openai#672) as the new target to beat - Reorders Top 8 directions: XSA-all confirmed at #1, Full GPTQ #2, SwiGLU #3, Muon-VS #4, aggressive quant #5, MASA openai#6, depth recurrence openai#7 with int6 risk warning, AdEMAMix openai#8 - Deprioritizes TTT-related directions already exploited by PR openai#672 - Collapses ~1000 lines of stale Round 0-3.9 session logs into a concise historical summary - Removes resolved blockers (flash_attn, SSH hangs, local runtime) - Adds fresh Round 1 section with 5 submitted experiments Co-Authored-By: Claude Opus 4.6 (1M context) <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
2 participants
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
Changes
records/track_non_record_16mb/2026-03-18_M3_DepthRecurrence_SwiGLU/train_gpt_mlx.pyrecords/track_non_record_16mb/2026-03-18_M3_DepthRecurrence_SwiGLU/submission.jsonrecords/track_non_record_16mb/2026-03-18_M3_DepthRecurrence_SwiGLU/README.mdLimitations
Trained on consumer hardware (M3/8GB) — score reflects hardware constraints, not the approach's ceiling. The same script can be run on 8×H100 for competitive results.