Add submission: Mixed Quantization + BigramHash + SWA (val_bpb 1.2421)#370
Open
SergheiBrinza wants to merge 3 commits intoopenai:mainfrom
Open
Add submission: Mixed Quantization + BigramHash + SWA (val_bpb 1.2421)#370SergheiBrinza wants to merge 3 commits intoopenai:mainfrom
SergheiBrinza wants to merge 3 commits intoopenai:mainfrom
Conversation
Personal case-study of my participation in the OpenAI Model Craft Challenge, plus the April Turbo-Muon submission brought to main so internal links resolve. Contents: - README.md: personal narrative and results tables - docs/METHODS.md: technical breakdown of each technique used - docs/EXPERIMENTS.md: verified runs and post-mortem of 020_ultimate - docs/UPSTREAM_README.md: original OpenAI README preserved for context - scripts/plot_curves.py: build training curves from train_*.log - assets/loss_curves.png: training dynamics of both submissions - Rewritten README for the 2026-03-21 submission - Full 2026-04-01 Turbo-Muon submission ported from the PR branch: README, submission.json, train_gpt.py, three seed logs Results on main: - 2026-03-21 Mixed Quantization + BigramHash + SWA: val_bpb 1.2421 - 2026-04-01 Turbo-Muon + EngramLite (3 seeds, std 0.0007): val_bpb 1.1431 Upstream PRs: - openai#370 - openai#1205
Removed 19 submission folders inherited from upstream fork since they belong to other participants and crowd the personal case-study. Only my two submissions remain under records/track_10min_16mb/. Also fixed one misattribution in docs/METHODS.md: the depth recurrence observation came from another participant's non-record experiment, not from my own run.
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.
10-layer Transformer with ReLU² MLP 3x, BigramHash embeddings, mixed INT6 quantization with STE, Stochastic Weight Averaging, and zstd-22 compression.