efficient_ai_CS6013_2026Fall

Evaluate a submission

python evaluation/eval_submission.py \
  --submission sample_submission/code.py \
  --model Qwen/Qwen3-4B-Instruct-2507

This compresses the model with the submission, restores it to HF format, then runs GPQA Diamond and MMLU-Pro (default: 10 samples each). Use --limit 1 for a smoke test.