openpsi-project/ReaLHF

Issues

• 加群二维码过期了

  #86 opened 21 days ago by LUMO666
  2

• grpo has not prm

  #79 opened 3 months ago by yiyepiaoling0715
  8

• 关于代码中数据传输、参数传输等的若干疑问

  #82 opened 2 months ago by metaqiang
  4

• How to debug remote subprocess?

  #81 opened 2 months ago by metaqiang
  5

• Suggestion for Fine-Grained Batch Control e.g `per_device_train_batch_size` or `mini_batch_size`

  #80 opened 3 months ago by dechunwang
  1

• 有多机分布式任务的example吗？

  #24 opened 3 months ago by PKUFlyingPig
  7

• DeepSpeed v0.14.0 is not compatible with the latest Nvidia PyTorch docker image.

  #77 opened 3 months ago by garrett4wade
  0

• Ray model workers may not be scheduled in a packed manner.

  #75 opened 3 months ago by garrett4wade
  0

• Automatic importing can be problematic if the parent directory contains "realhf".

  #58 opened 3 months ago by garrett4wade
  0

• Is CUDAGraph generation available during PPO training time？

  #70 opened 3 months ago by DZ9
  5

• allocation_mode

  #45 opened 4 months ago by AIRobotZhang
  2

• TransformerConfig() takes no arguments

  #46 opened 4 months ago by AIRobotZhang
  2

• Issues for the generation experiment.

  #59 opened 4 months ago by garrett4wade
  0

• Error while deserializing header: MetadataIncompleteBuffer

  #44 opened 5 months ago by AIRobotZhang
  2

• No module named 'colorlog'

  #41 opened 5 months ago by AIRobotZhang
  1

• 如何推理

  #40 opened 5 months ago by AIRobotZhang
  2

• worker ERROR: Worker encountered error 'id'

  #43 opened 5 months ago by AIRobotZhang
  1

• Loading parameters takes exceptionally long time.

  #18 opened 6 months ago by nuzant
  1

• Rewards do not increase with `use_cuda_graph=True`.

  #14 opened 6 months ago by garrett4wade
  1

• 你们论文中的实验有认真调过其他框架的性能么？

  #10 opened 6 months ago by hijkzzz
  3