Skip to content

feat: add local weight synchronization to CheckpointEngineManager - #276

Merged
kevssim merged 1 commit into
modelscope:dev-swift-v5from
kevssim:dev-ckpt-engine
Aug 28, 2026
Merged

feat: add local weight synchronization to CheckpointEngineManager#276
kevssim merged 1 commit into
modelscope:dev-swift-v5from
kevssim:dev-ckpt-engine

Conversation

@kevssim

@kevssim kevssim commented Aug 28, 2026

Copy link
Copy Markdown
Collaborator

PR type

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

PR information

Write the detail information belongs to this PR.

  • Introduce mode: auto, naive, colocate, and standalone.
  • In naive mode, local model and sampler objects synchronize weights directly through a generator without creating a checkpoint engine.
  • Keep CUDA IPC for explicit colocate Ray deployments and NCCL/HCCL for standalone deployments.
  • auto selects naive for local objects and standalone for Ray actor handlers.
  • Update Transformers/Megatron models and vLLM/SGLang samplers for direct generator-based synchronization.

@tastelikefeet

Copy link
Copy Markdown
Collaborator

Now the branch is a draft, you can directly merge into it.

@kevssim

kevssim commented Aug 28, 2026

Copy link
Copy Markdown
Collaborator Author

Now the branch is a draft, you can directly merge into it.现在这个分支是草稿状态,你可以直接合并进去。

我还需要对代码进行一些测试和修改,完成后合入

@kevssim
kevssim marked this pull request as ready for review August 28, 2026 03:33
@kevssim
kevssim merged commit 6ec8250 into modelscope:dev-swift-v5 Aug 28, 2026
0 of 3 checks passed
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