Repository Issue Activity (beta)

#124 Verify evals on Papers with Code

#69 train/train_rollout_logprob_abs_diff 值过大

#116 把OPD加到优势计算后，训练过程中：entropy和kl_loss不断增加，response_len不断变短，reward不断降低！这是什么原因，有遇到过吗？

#118 Is there anyone experienced who could try running this operating-system-like project? I can provide the .md file, but I’m not getting it to work properly on my end.

#104 是否有qwen3 8B 在terminal rl 上的训练曲线？

#62 tool call training question

#113 关于 Personal Agent GSM8K 实验中 Student/Teacher 场景训练方式

#94 Qwen3.5-4B无法用于训练

#111 Docker compose command failed with exit code 17

#112 是否有Qwen3-4B LoRA版本在Personal Agent Track上的训练结果？

#58 slime以及megatron版本的问题

#75 toolcall 程序hang了，求助~

#107 Clarification about the pipeline

#88 rollout阶段很慢

#101 openclaw-opd/4b_lora训练报错问题

#98 想知道这个项目支持什么显卡？

#92 SWE-RL 使用到的镜像失效了，请问这些镜像需要什么，如果用自定义镜像的话需要保证镜像有什么功能吗？

#95 transformer_engine 装不上

#91 Possible complementary direction: post-task skill lifecycle management with SkillClaw

#76 ray ipv6

#68 训练效果问题

#87 Dockfile 格式问题

#82 Great work on this project! I have a question regarding the current framework when running reinforcement learning training in OSWorld. Is it possible to use my own CPU server to run the Docker environment instead of relying on a cloud server?

#83 What PRM model was used for the personal-agent experiments, and how were teacher_log_probs obtained in OPD?

#84 torch.AcceleratorError: CUDA error: an illegal memory access was encountered