Issues

• Policy drift from minibatch updates vs. Smaller rollout batches

  BytedTsinghua-SIA/DAPO#40 · Filter by author SangYeop-Lee opened on Mar 6, 2026

• Missing of the scripts of Training and Reproducibility

  BytedTsinghua-SIA/DAPO#39 · Filter by author Wizard759 opened on Aug 26, 2025
  2

• why the reuslts of eval_aime24.py is acc/mean@32

  BytedTsinghua-SIA/DAPO#38 · Filter by author baoshenghe0321 opened on Jul 28, 2025

• question on training wandb

  BytedTsinghua-SIA/DAPO#37 · Filter by author DBtxy opened on Jul 4, 2025

• size of math_dapo dataset

  BytedTsinghua-SIA/DAPO#36 · Filter by author BinceQu opened on Jun 27, 2025
  1

• How can I pin the GPU memory?

  BytedTsinghua-SIA/DAPO#35 · Filter by author congli01 opened on Jun 24, 2025

• Old policy update frequency

  BytedTsinghua-SIA/DAPO#34 · Filter by author LiGuo12 opened on Jun 21, 2025

• RL method used in section 3.1 and 3.3

  BytedTsinghua-SIA/DAPO#33 · Filter by author Riezmann75 opened on Jun 17, 2025

• reproduce dapo with ValueError: Total available GPUs 4.0 is less than total desired GPUs 32

  BytedTsinghua-SIA/DAPO#32 · Filter by author chuangzhidan opened on May 24, 2025

• clip_higher 的验证逻辑是否有问题？

  BytedTsinghua-SIA/DAPO#31 · Filter by author harmonytan opened on May 11, 2025
  1

• License about the dataset

  BytedTsinghua-SIA/DAPO#30 · Filter by author DyeKuu opened on May 2, 2025

• Could you please share the DAPO-Math-17K dataset without Dataset Transformation applied?

  BytedTsinghua-SIA/DAPO#29 · Filter by author carrot0117 opened on Apr 27, 2025

• Language Consistency

  BytedTsinghua-SIA/DAPO#27 · Filter by author BeastyZ opened on Apr 13, 2025

• The actor died because of an error raised in its creation task, ray::upgRRXWorkerDict_0:1:WorkerDict.__init__()

  BytedTsinghua-SIA/DAPO#26 · Filter by author SharonJin422 opened on Apr 11, 2025
  1

• 您·好 我想请问 dapo中的Overlong Filtering和Soft overlong punishment

  BytedTsinghua-SIA/DAPO#25 · Filter by author Zero-coder opened on Apr 9, 2025
  4

• Question about Figure 4(a) and Token-Level Policy Gradient Loss

  BytedTsinghua-SIA/DAPO#24 · Filter by author carrot0117 opened on Apr 8, 2025

• [question about the paper]

  BytedTsinghua-SIA/DAPO#21 · Filter by author EmmittXu opened on Apr 7, 2025

• [Question] which version of huggingface hub is required? or any scripts to prepare & download Qwen2.5-32B to /root/verl/models/Qwen2.5-32B

  BytedTsinghua-SIA/DAPO#20 · Filter by author blackyang opened on Mar 29, 2025

• Question about equation 4 in the paper

  BytedTsinghua-SIA/DAPO#17 · Filter by author naajeehxe opened on Mar 25, 2025

• What is the maximum prompt length of the DAPO-Math-17k dataset?

  BytedTsinghua-SIA/DAPO#14 · Filter by author zhengkw18 opened on Mar 24, 2025
  1

• Ablation on removing KL term

  BytedTsinghua-SIA/DAPO#9 · Filter by author zizhao-hu opened on Mar 21, 2025
  2

• can also bring improvements to smaller models, such as those with 7 billion parameters?"

  BytedTsinghua-SIA/DAPO#7 · Filter by author qiuwenbogdut opened on Mar 19, 2025
  1

• Several questions about the paper

  BytedTsinghua-SIA/DAPO#6 · Filter by author tqzhong opened on Mar 19, 2025
  1

• [Question] base-model selection suggest

  BytedTsinghua-SIA/DAPO#2 · Filter by author jszyxw00 opened on Mar 18, 2025
  1

• what is the minimum number of gpus needed to train this model?

  BytedTsinghua-SIA/DAPO#1 · Filter by author yxchng opened on Mar 18, 2025
  3