[RL] [FDConfig] Disable overlap schedule by default in RL#7739
[RL] [FDConfig] Disable overlap schedule by default in RL#7739liyonghua0910 wants to merge 1 commit intoPaddlePaddle:developfrom
Conversation
|
Thanks for your contribution! |
eb38682 to
7edb89a
Compare
PaddlePaddle-bot
left a comment
There was a problem hiding this comment.
🤖 Paddle-CI-Agent | pr_review |
2026-05-07 19:19:30
📋 Review 摘要
PR 概述:在 FDConfig.postprocess() 中新增保护逻辑,当 dynamic_load_weight=True(RL 场景)时强制禁用 overlap schedule,规避已知 Bug。
变更范围:fastdeploy/config.py
影响面 Tag:[FDConfig] [RL]
📝 PR 规范检查
PR 标题中的 [FDConfig] Tag 合规;但 PR 描述中所有 section 内容均为空/仅模板占位符,不合规,需要填充实际内容。
标题建议(可直接复制):
[FDConfig][RL] Disable overlap schedule when dynamic_load_weight is enabled
PR 描述建议(可直接复制,必须复刻 checklist §D2 模板的完整结构):
## Motivation
在 RL 训练场景下(`dynamic_load_weight=True`),overlap schedule 存在已知 Bug,会导致运行异常。当前临时在 `FDConfig.postprocess()` 中强制禁用,待 Bug 修复后再恢复启用。
## Modifications
- `fastdeploy/config.py`:在 `FDConfig.postprocess()` 中新增判断逻辑:当 `load_config` 不为 None 且 `load_config.dynamic_load_weight is True` 时,将 `scheduler_config.enable_overlap_schedule` 强制设为 `False`。
## Usage or Command
N/A
## Accuracy Tests
N/A
## Checklist
- [x] 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.
- [x] Add unit tests. Please write the reason in this PR if no unit tests.
- [x] 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.问题
未发现阻塞性问题。
总体评价
变更逻辑清晰,在 postprocess 中统一处理 RL 场景下的 overlap schedule 禁用,是合理的防御性处理。PR 描述需要补充完整,建议同步添加 [RL] Tag 以体现实际影响面。
CI报告基于以下代码生成(30分钟更新一次): 1 任务总览所有 Required 任务均已通过,建议合并。✅
2 任务状态汇总2.1 Required任务 : 2/2 通过
2.2 可选任务 — 9/15 通过
3 失败详情(仅 required)无 required 失败任务。 |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## develop #7739 +/- ##
==========================================
Coverage ? 71.62%
==========================================
Files ? 396
Lines ? 55571
Branches ? 8689
==========================================
Hits ? 39803
Misses ? 13029
Partials ? 2739
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Motivation
暂时针对 RL 场景 (dynamic_load_weight=True) 关闭 overlap schedule,已知问题修复后再重新打开。
Modifications
Usage or Command
Accuracy Tests
Checklist
[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]]pre-commitbefore commit.releasebranch, make sure the PR has been submitted to thedevelopbranch, then cherry-pick it to thereleasebranch with the[Cherry-Pick]PR tag.