Skip to content

feat: consolidate moonshot stack into clean submission-ready training script #7

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
Copilot wants to merge 5 commits into
base: main
Choose a base branch
from
+2,103 −885
Draft

feat: consolidate moonshot stack into clean submission-ready training script #7

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
8bb32af
Initial plan
Copilot Apr 3, 2026
376b8e0
feat: clean submission-ready training script with full moonshot stack
Copilot Apr 3, 2026
8793084
Initial plan
Copilot Apr 4, 2026
36f0109
Trim train_gpt_mlx_kl.py from 1847 to 1493 lines, fix clip_grad_tree bug
Copilot Apr 4, 2026
5c2050e
Merge pull request #8 from kailean/copilot/verify-aub-1-0-bpb-code-le…
kailean Apr 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions CLAUDE.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,9 +5,11 @@ OpenAI competition to train the best LM that fits in 16MB.
Baseline: train_gpt_mlx.py (stock), my version: train_gpt_mlx_kl.py

## Commands
- Smoke test: RUN_ID=test ITERATIONS=100 TRAIN_BATCH_TOKENS=8192 VAL_LOSS_EVERY=0 VAL_BATCH_SIZE=8192 WARMUP_STEPS=3 python3 train_gpt_mlx_kl.py
- Must activate venv first: source ~/pg_env/bin/activate (venv is at ~/pg_env, NOT .venv)
- Smoke test (baseline): `RUN_ID=test ITERATIONS=100 TRAIN_BATCH_TOKENS=8192 VAL_LOSS_EVERY=0 VAL_BATCH_SIZE=8192 WARMUP_STEPS=3 python3 train_gpt_mlx_kl.py`
- Smoke test (moonshot): `RUN_ID=test ITERATIONS=100 TRAIN_BATCH_TOKENS=8192 VAL_LOSS_EVERY=0 VAL_BATCH_SIZE=8192 WARMUP_STEPS=3 ENGRAM_LITE_ENABLED=1 COMPLEMENT_ALPHA=0.5 NGRAM_MIXER_ENABLED=1 python3 train_gpt_mlx_kl.py`
- Must activate venv first: `source ~/pg_env/bin/activate` (venv is at ~/pg_env, NOT .venv)
- Data is in ./data/datasets/fineweb10B_sp1024/
- Full moonshot run (8×H100): `ENGRAM_LITE_ENABLED=1 COMPLEMENT_ALPHA=0.5 NGRAM_MIXER_ENABLED=1 NGRAM_ALPHA=0.25 NGRAM_MAX_ORDER=4 python3 train_gpt_mlx_kl.py`

## Key metrics
- train_loss: lower is better, compare at same step count
Expand Down
Loading