[BugFix]fix chunk_prefill logprobs

[bukejiyu]

Motivation

💡 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)

修复开启chunked prefill 跑logprobs 时 异常挂掉的问题
当开启chunk_prefill的时候 同一个req0 可能被分成了 3个chunked [chunked0,chunke1,chunke2]
step=0的时候 调度传来chunked0
step=1的时候 调度传来 chunked1
但是可能在step=2的时候调度并没有传来 chunke2
导致计算prompt logprobs的时候会用 req0的信息和不带req0的 hidden_state
导致计算挂掉
该PR 通过判断 req0对应的槽位index 和 seq_this_time[index] 是否大于0判断
是否存在需要处理的prompt_hidden_state以此来修复切chunked prompt logprob挂掉的问题

Modifications

Usage or Command

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.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 33.33333% with 2 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@aa50bd2). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
fastdeploy/worker/gpu_model_runner.py	33.33%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #8087   +/-   ##
==========================================
  Coverage           ?   67.79%           
==========================================
  Files              ?      475           
  Lines              ?    66915           
  Branches           ?    10321           
==========================================
  Hits               ?    45363           
  Misses             ?    18659           
  Partials           ?     2893           

Flag	Coverage Δ
GPU	77.85% <33.33%> (?)
XPU	6.95% <0.00%> (?)

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

☔ View full report in Codecov by Harness.
📢 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.

[PaddlePaddle-bot]

🤖 Paddle-CI-Agent | pr_review | 2026-07-01 17:06:29

📋 Review 摘要

PR 概述：为 GPU prompt logprobs 在 chunked prefill 场景增加 current-step 过滤，避免处理本步未执行的 pending 请求。
变更范围：fastdeploy/worker/gpu_model_runner.py、tests/worker/test_gpu_prompt_logprobs.py
影响面 Tag：[Engine]

问题

级别	文件	概述
🟡 建议	tests/worker/test_gpu_prompt_logprobs.py:176	新测试仍只覆盖当前请求 seq_lens_this_time_cpu > 0 的正常计算分支，未覆盖本次新增的 skip 条件

历史 Findings 修复情况

Finding	问题	状态
F1	GPU prompt logprobs 的 current-step 检查仍使用未随 reorder 对齐的 req_ids.index(req_id)	⚠️ 仍存在
F2	同构的 XPU/Metax prompt logprobs 路径仍未同步 current-step 过滤	⚠️ 仍存在

📝 PR 规范检查

标题缺少 ] 后的空格，且 PR 描述保留了模板占位内容，未填写 Motivation/Modifications/Usage/Accuracy Tests 的具体内容。

标题建议（可直接复制）：

• [BugFix] Fix chunked prefill prompt logprobs

PR 描述建议（点击展开，可直接复制）

## Motivation
Fix incorrect prompt logprobs handling in chunked prefill when pending prompt_logprobs requests are not part of the current forward step.

## Modifications
- In `fastdeploy/worker/gpu_model_runner.py::_get_prompt_logprobs_list`, skip prompt_logprobs requests whose request id is not in the current `share_inputs["req_ids"]` or whose `seq_lens_this_time_cpu` is 0.
- This prevents computing prompt logprobs from hidden states of requests that did not run in the current step.

## 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.
- [ ] 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.

总体评价

当前修改方向对准了 chunked prefill 下 prompt logprobs 的 pending request 问题，但新增测试尚未守住这次修复的跳过分支。历史遗留的 slot 对齐与多硬件同步问题仍未从当前 diff 中看到修复。

[PaddlePaddle-bot]

🟡 建议 这个测试把 req_ids 设成当前请求并将 seq_lens_this_time_cpu[0] 设为 4，因此仍只走正常计算分支；PR 新增的 idx < 0 / seq_lens_this_time_cpu <= 0 skip 分支没有被验证。

建议补一个用例：在 prompt_logprobs_reqs 中放入一个本轮未执行的请求（req_id 不在 share_inputs["req_ids"]，或对应 seq_lens_this_time_cpu 为 0），并断言 _get_prompt_logprobs_list() 不调用 compute_logits、不写入该请求的 in_progress_prompt_logprobs，返回列表保持 None。

[paddle-bot]

Thanks for your contribution!