Skip to content

Conversation

@hjh0119
Copy link
Collaborator

@hjh0119 hjh0119 commented Mar 25, 2025

PR type

  • Bug Fix
  • New Feature
  • Document Updates
  • More Models or Datasets Support

PR information

reset prefix cache after vllm model weight update

Experiment results

Paste your experiment result here(if needed).

@hjh0119 hjh0119 merged commit f6cf96c into modelscope:main Mar 25, 2025
1 of 2 checks passed
@hjh0119 hjh0119 deleted the reset-prefix-cache branch March 25, 2025 03:35
tastelikefeet added a commit to tastelikefeet/swift that referenced this pull request Mar 25, 2025
…0325

* commit '20cfa3396705a50230d5cb64850e54ba5043ee2c': (23 commits)
  compat vllm0.8.1 (modelscope#3656)
  fix grpo vllm tp (modelscope#3658)
  fix label_names (modelscope#3657)
  set grpo multi turn max tokens (modelscope#3655)
  update docs (modelscope#3653)
  fix grpo epsilon(modelscope#3652)
  Fix template torch_dtype (modelscope#3651)
  [grpo] separate the epsilon (modelscope#3599)
  fix grpo pt ddp (modelscope#3648)
  fix prm (modelscope#3647)
  grpo reset prefix cache (modelscope#3640)
  fix grpo warning (modelscope#3630)
  support qwen2_5_vl_32b (modelscope#3642)
  fix reward model (modelscope#3641)
  fix grpo lora split module (modelscope#3635)
  fix grpo cosine reward (modelscope#3638)
  Support deepseek v3 0324 (modelscope#3637)
  update docs (modelscope#3633)
  fix (modelscope#3632)
  support train_sampler_random (modelscope#3631)
  ...

# Conflicts:
#	docs/source/Instruction/GRPO.md
#	docs/source/Instruction/命令行参数.md
#	docs/source_en/Instruction/Command-line-parameters.md
#	docs/source_en/Instruction/GRPO.md
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants