Issues

• Are prosqa_cot.yaml and prontoqa_cot.yaml available?

  facebookresearch/coconut#57 · Filter by author SMU-Rainflow opened on May 15, 2026

• What is the value of uniform_prob when reproducing?

  facebookresearch/coconut#54 · Filter by author fatRobert23361 opened on Mar 30, 2026
  1

• Normalize the last hidden state

  facebookresearch/coconut#51 · Filter by author KevinZhoutianyi opened on Dec 22, 2025

• Training Instability in COCONUT Latent Reasoning Stage

  facebookresearch/coconut#50 · Filter by author gdgdandsz opened on Dec 9, 2025
  2

• Won't forwarding the model autoregressively multiple times during training consume numerous VRAM?

  facebookresearch/coconut#48 · Filter by author Darkbblue opened on Dec 2, 2025
  1

• Questions regarding performance on GSM8kaug

  facebookresearch/coconut#45 · Filter by author mfahes opened on Oct 15, 2025
  1

• Number of epochs in initial stage for gsm8k

  facebookresearch/coconut#43 · Filter by author connordilgren opened on Sep 20, 2025
  1

• Meta gradients and RL for adaptable simultaneous reasoning

  facebookresearch/coconut#42 · Filter by author arccoxx opened on Aug 15, 2025

• Optimizer is reset every epoch rather than during training stage switch

  facebookresearch/coconut#39 · Filter by author JaydenTeoh opened on Jul 2, 2025
  1

• Will you consider releasing some checkpoints?

  facebookresearch/coconut#37 · Filter by author Mike001-wq opened on Jun 5, 2025
  1

• why MyCollator left pad and right pad the features?

  facebookresearch/coconut#33 · Filter by author jasonrichdarmawan opened on May 13, 2025
  2

• latent token and CoT match = 0

  facebookresearch/coconut#31 · Filter by author Sumireko-Usami opened on Apr 18, 2025
  3

• Great idea.

  facebookresearch/coconut#28 · Filter by author BlackTea-c opened on Apr 8, 2025
  3

• gsm dataset training phase coconut replacing each stage c thoughts

  facebookresearch/coconut#27 · Filter by author sjfievxcjkfd2 opened on Apr 1, 2025
  1

• Support for Multi-Node Multi-GPU Training

  facebookresearch/coconut#26 · Filter by author Geaming2002 opened on Mar 29, 2025
  1

• Question about the best model performance at the 6th epoch

  facebookresearch/coconut#25 · Filter by author FYYDCC opened on Mar 21, 2025
  1

• Errors when trying to train deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B with COCONUT method

  facebookresearch/coconut#24 · Filter by author NerdyVisky opened on Mar 19, 2025
  2

• Adding configuration to run on a single A100 GPU

  facebookresearch/coconut#23 · Filter by author NerdyVisky opened on Mar 19, 2025
  1

• 数据集构建失败

  facebookresearch/coconut#21 · Filter by author sjfievxcjkfd2 opened on Mar 16, 2025
  1

• coconut w/o curriculum

  facebookresearch/coconut#20 · Filter by author PolinaDruzhinina opened on Mar 10, 2025
  3

• more thoughts

  facebookresearch/coconut#17 · Filter by author dox012 opened on Mar 6, 2025
  1

• loss increasing

  facebookresearch/coconut#16 · Filter by author dox012 opened on Mar 6, 2025
  3