Issues

• LLaMA-3.2-1B-Instruct math results diverge significantly from paper (GSM8k-Aug, GSM8k-Aug-NL)

  zhenyi4/codi#17 · Filter by author andrewivan123 opened on Apr 15, 2026
  1

• attn_to_lats not populated

  zhenyi4/codi#16 · Filter by author neverix opened on Jan 11, 2026

• 关于HF检查点的询问

  zhenyi4/codi#15 · Filter by author Frosty-Heart opened on Jan 2, 2026
  1

• Request for the hyperparameters of SFT base

  zhenyi4/codi#14 · Filter by author scarydemon2 opened on Dec 26, 2025

• Why the model weights from huggingface can't be loaded in the scripts/test_gpt2.sh

  zhenyi4/codi#13 · Filter by author hzy051302 opened on Dec 11, 2025
  1

• About the reproduction issue.

  zhenyi4/codi#12 · Filter by author lqcStar opened on Dec 8, 2025
  2

• Minor typo in Appendix A (Coconut on LLaMA-1B)

  zhenyi4/codi#8 · Filter by author J1mL1 opened on Nov 28, 2025
  1

• Why is the model called twice here to generate ref_outputs, which are then used separately for the cross-entropy loss and the distillation loss?

  zhenyi4/codi#7 · Filter by author whitesweater opened on Nov 4, 2025
  1

• it doesn't work when using qwen3

  zhenyi4/codi#6 · Filter by author hzy051302 opened on Oct 29, 2025

• loss not anumber

  zhenyi4/codi#5 · Filter by author livehappy1 opened on Oct 22, 2025

• why the training scripts do not set the fix_attn_mask to True?

  zhenyi4/codi#4 · Filter by author BIGKnight opened on Oct 5, 2025
  1

• the last cot step

  zhenyi4/codi#3 · Filter by author myla01 opened on Sep 1, 2025
  1