volcengine/verl

Issues

• Question about actor training-rollout resharding

  #73 opened 4 days ago by 0oshowero0
  2

• Support RLOO/GRPO/REINFORCE?

  #68 opened 6 days ago by fzyzcjy
  24

• [ray] latest ray compatibility

  #46 opened 4 days ago by eric-haibin-lin
  1

• Do we have plans for data packing?

  #53 opened 21 days ago by YixinSong-e
  6

• Missing doc about "Algorithm Baselines"

  #66 opened 5 days ago by fzyzcjy
  2

• does this framework support long-generation such 8k-16k

  #69 opened 6 days ago by yyht
  1

• Actor model didn't update correctly when upgrade megatron to core-r0.6.0

  #64 opened 13 days ago by Wodswos
  1

• package confilct

  #62 opened 15 days ago by hljjjmssyh
  1

• Are optimizer states reloaded or offloaded during the conversion from actor training to actor rollout?

  #42 opened 20 days ago by G1aZzz
  1

• Question about recomputation in actor module

  #51 opened 21 days ago by 0oshowero0
  3

• Calling for Improving Robustness of FSDP-vLLM Rollout

  #48 opened 22 days ago by nwiad
  4

• Hangs during vllm rollout, no error message

  #12 opened 2 months ago by Vamix
  5

• whether the auto device mapping code in the paper has been uploaded?

  #5 opened 23 days ago by Zeroreoo
  2

• Docker image support

  #8 opened a month ago by SolenoidWGT
  3

• Does this framework support full parameter PPO tuning for the Qwen2.5-14B model on 8-A100 GPUs with 80GB memory each?

  #40 opened a month ago by hljjjmssyh
  1

• Questions Regarding Generation Weights Offloading and Buffer Usage

  #25 opened a month ago by metaqiang
  1

• enable_gradient_checkpointing is not working

  #26 opened a month ago by Vamix
  2

• Why create_colocated_worker_cls and spawn

  #29 opened a month ago by eelxpeng
  2

• Basic Tutorial: Adding a New LLM Inference/Serving Backend

  #21 opened 2 months ago by PeterSH6
  1

• Unexpected Increase in Rollout Time After Reducing num_hidden_layers in deepseek-llm-7b-chat Model

  #24 opened a month ago by metaqiang
  2

• Why the `magatron_v4.patch` is needed?

  #14 opened 2 months ago by hxdtest
  4

• Is non-RmPad version model and RmPad verison mdoel interchangeable?

  #20 opened 2 months ago by yanggthomas
  5

• [Roadmap] veRL Development Roadmap

  #22 opened 2 months ago by PeterSH6
  0

• 关于数据和参数切分的性能测试问题

  #16 opened 2 months ago by metaqiang
  4

• 有提供性能调试的手段吗？

  #11 opened 2 months ago by metaqiang
  14

• [RFC] Megatron-LM and MCore maintaining issues for veRL

  #15 opened 2 months ago by PeterSH6
  0

• KeyError: 'raw_prompt'

  #13 opened 2 months ago by YixinSong-e
  2

• 启动训练脚本出现偶发性ray.exceptions.ActorDiedError错误

  #10 opened 2 months ago by metaqiang
  2

• Can I run ppo in llama3.1-70B-instruct?

  #6 opened 2 months ago by cingtiye
  1