Skip to content

[Cherry-Pick][RL] Adapt async rollout checkpoint update flow(#7042)#7084

Merged
Jiang-Jia-Jun merged 1 commit intoPaddlePaddle:release/2.5from
jackyYang6:cp/rlhf-async
Mar 31, 2026
Merged

[Cherry-Pick][RL] Adapt async rollout checkpoint update flow(#7042)#7084
Jiang-Jia-Jun merged 1 commit intoPaddlePaddle:release/2.5from
jackyYang6:cp/rlhf-async

Conversation

@jackyYang6
Copy link
Copy Markdown
Contributor

Motivation

This PR cherry-picks #7042 to release/2.5.

It aligns the async rollout checkpoint update flow with develop, including the update_weights parameter update from rsync_config to verify_checksum, and keeps the route validation test in sync.

💡 If this PR is a Cherry Pick, the PR title needs to follow the format by adding the [Cherry-Pick] label at the very beginning and appending the original PR ID at the end. For example, [Cherry-Pick][CI] Add check trigger and logic(#5191)

💡 如若此PR是Cherry Pick,PR标题需遵循格式,在最开始加上[Cherry-Pick]标签,以及最后面加上原PR ID,例如[Cherry-Pick][CI] Add check trigger and logic(#5191)

Modifications

  • Align update_weights parameter passing in the RL async update flow
  • Adapt checkpoint verification parameter propagation across:
    • api_server
    • gpu_worker
    • gpu_model_runner
    • dynamic_weight_manager
  • Sync the same control-path adaptation for Metax worker/model runner
  • Fix cross-layer interface mismatch in async rollout checkpoint updates

Usage or Command

Example control request used in RL async integration:

curl http://<host>:<port>/v1/update_weights \
  -H "Content-Type: application/json" \
  -d '{
    "version": "0",
    "verify_checksum": false
  }'

Typical control flow during async rollout:

pause -> update_weights -> resume

Accuracy Tests

Checklist

  • Add at least a tag in the PR title.
    • Tag list: [[FDConfig],[APIServer],[Engine], [Scheduler], [PD Disaggregation], [Executor], [Graph Optimization], [Speculative Decoding], [RL], [Models], [Quantization], [Loader], [OP], [KVCache], [DataProcessor], [BugFix], [Docs], [CI], [Optimization], [Feature], [Benchmark], [Others], [XPU], [HPU], [GCU], [DCU], [Iluvatar], [Metax]]
    • You can add new tags based on the PR content, but the semantics must be clear.
  • Format your code, run pre-commit before commit.
  • Add unit tests. Please write the reason in this PR if no unit tests.
  • Provide accuracy results.
  • If the current PR is submitting to the release branch, make sure the PR has been submitted to the develop branch, then cherry-pick it to the release branch with the [Cherry-Pick] PR tag.

@paddle-bot
Copy link
Copy Markdown

paddle-bot Bot commented Mar 30, 2026

Thanks for your contribution!

* update checkpoint-transfer flow and control update_weights params

* test: add update_weights route validation

(cherry picked from commit 05f2d95)
@codecov-commenter
Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 28.00000% with 18 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (release/2.5@971fc7c). Learn more about missing BASE report.

Files with missing lines Patch % Lines
fastdeploy/rl/dynamic_weight_manager.py 6.25% 15 Missing ⚠️
fastdeploy/entrypoints/openai/api_server.py 66.66% 0 Missing and 1 partial ⚠️
fastdeploy/worker/gpu_model_runner.py 66.66% 1 Missing ⚠️
fastdeploy/worker/gpu_worker.py 66.66% 1 Missing ⚠️
Additional details and impacted files
@@              Coverage Diff               @@
##             release/2.5    #7084   +/-   ##
==============================================
  Coverage               ?   69.03%           
==============================================
  Files                  ?      390           
  Lines                  ?    54077           
  Branches               ?     8515           
==============================================
  Hits                   ?    37334           
  Misses                 ?    14039           
  Partials               ?     2704           
Flag Coverage Δ
GPU 69.03% <28.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@Jiang-Jia-Jun Jiang-Jia-Jun merged commit 077ec83 into PaddlePaddle:release/2.5 Mar 31, 2026
33 of 38 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.

3 participants