Search results

Normalize the last hidden state

Status: Open.

#51 In facebookresearch/coconut;

· KevinZhoutianyi opened on Dec 22, 2025

Training Instability in COCONUT Latent Reasoning Stage

Status: Open.

#50 In facebookresearch/coconut;

· gdgdandsz opened on Dec 9, 2025

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

Status: Open.

#48 In facebookresearch/coconut;

· Darkbblue opened on Dec 2, 2025

Questions regarding performance on GSM8kaug

Status: Open.

#45 In facebookresearch/coconut;

· mfahes opened on Oct 15, 2025

Number of epochs in initial stage for gsm8k

Status: Open.

#43 In facebookresearch/coconut;

· connordilgren opened on Sep 20, 2025

Meta gradients and RL for adaptable simultaneous reasoning

Status: Open.

#42 In facebookresearch/coconut;

· arccoxx opened on Aug 15, 2025

Optimizer is reset every epoch rather than during training stage switch

Status: Open.

#39 In facebookresearch/coconut;

· JaydenTeoh opened on Jul 2, 2025

Will you consider releasing some checkpoints?

Status: Open.

#37 In facebookresearch/coconut;

· Mike001-wq opened on Jun 5, 2025

why MyCollator left pad and right pad the features?

Status: Open.

#33 In facebookresearch/coconut;

· jasonrichdarmawan opened on May 13, 2025

latent token and CoT match = 0

Status: Open.

#31 In facebookresearch/coconut;

· Sumireko-Usami opened on Apr 18, 2025

Great idea.

Status: Open.

#28 In facebookresearch/coconut;

· BlackTea-c opened on Apr 8, 2025

gsm dataset training phase coconut replacing each stage c thoughts

Status: Open.

#27 In facebookresearch/coconut;

· sjfievxcjkfd2 opened on Apr 1, 2025