[Others] update flash mask version

[BingooYang]

Motivation

flashinfer升级到0.6.11版本要求nvidia-cutlass-dsl>=4.4.2（https://github.com/PaddlePaddle/FastDeploy/pull/7799），flash mask旧版本锁死了nvidia-cutlass-dsl==4.4.2，产生冲突，因此升级一下flash mask版本

Modifications

升级flash mask版本
版本信息记录在：https://ku.baidu-int.com/knowledge/HFVrC7hq1Q/pKzJfZczuc/YeqWcBGW4m/EUBpKxHfTurV5G

Usage or Command

NA

Accuracy Tests

NA

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]

Thanks for your contribution!

[PaddlePaddle-bot]

🤖 Paddle-CI-Agent | ci_status_monitor | 2026-05-20 03:21:58

CI报告基于以下代码生成（30分钟更新一次）:

• PR commit: e5d5437
• Merge base: dad5a43 (branch: develop)
• 查看完整 Diff
• CI 详情

---

1 任务总览

有 2 个 required 任务失败，需优先处理后方可合并。

总执行（rerun次数）	总任务	✅ 通过	❌ 失败	⏳ 运行中	⏸️ 等待中	跳过
42(0)	42	38	3	0	1	0

---

2 任务状态汇总

2.1 Required任务 : 8/10 通过

必选任务阻塞合并，失败需优先处理。

状态	任务	耗时	根因	修复建议	日志	重跑
❌	Run FastDeploy Unit Tests and Coverage / run_tests_with_coverage	1h19m	PR问题：新增 flash_attn_backend.py 7行覆盖率 0%	为 L88-99 新增单测或申请豁免	Job	-
❌	Approval	8s	PR问题：新增 logger.info 触发日志修改审批规则	请 xyxinyang 或 zyyzghb Approve	Job	-
✅	其余 8 个必选任务通过	-	-	-	-	-

2.2 可选任务 — 30/32 通过

可选任务不阻塞合并，失败仅供参考。

状态	任务	耗时	日志	重跑
❌	Run iluvatar Tests / run_iluvatar_cases	1m47s	Job	-
⏸️	CI_HPU	-	-	-
✅	其余 30 个可选任务通过	-	-	-

---

3 失败详情（仅 required）

Run FastDeploy Unit Tests and Coverage / run_tests_with_coverage — 覆盖率不达标（置信度: 高）

Run FastDeploy Unit Tests and Coverage / run_tests_with_coverage

• 状态: ❌ 失败
• 错误类型: 覆盖率不达标
• 置信度: 高
• 根因摘要: PR 新增 flash_attn_backend.py 的7行逻辑覆盖率为 0%
• 分析器: ci_analyze_unittest_fastdeploy

失败用例: 无（单元测试全部通过，覆盖率检查失败）

根因详情:
PR 在 flash_attn_backend.py 第88-99行新增了 try/except 的 flash mask 条件导入逻辑，这些新增行（L88, L89, L91, L92, L96, L98, L99）没有被任何单元测试覆盖，diff 覆盖率为 0%（7/7 行均为 violation），低于 80% 的覆盖率阈值，触发 COVERAGE_EXIT_CODE=9，最终 exit code 9。注意单元测试本身全部通过（TEST_EXIT_CODE=0），仅覆盖率检查失败。

关键日志:

COVERAGE_EXIT_CODE: 9
GPU Patch Coverage Details:
{"src_stats": {"fastdeploy/model_executor/layers/attention/flash_attn_backend.py":
  {"percent_covered": 0.0, "violation_lines": [88, 89, 91, 92, 96, 98, 99],
   "covered_lines": []}},
 "total_num_lines": 7, "total_num_violations": 7, "total_percent_covered": 0}
##[error]Process completed with exit code 9.

修复建议:

1. 为 fastdeploy/model_executor/layers/attention/flash_attn_backend.py L88-99 新增 mock 单元测试，覆盖 is_flash_mask_available() 返回 True/False 两个分支，以及 ImportError/ModuleNotFoundError 异常分支
2. 若 CI 环境无法模拟 GPU 相关 import，可在 CI 配置中为该文件的条件导入代码申请覆盖率豁免

修复建议摘要: 为 flash_attn_backend.py L88-99 新增单测或申请豁免

关联变更: fastdeploy/model_executor/layers/attention/flash_attn_backend.py L88-99（新增 flash mask 条件导入 try/except 逻辑）
链接: 查看日志

Approval — 代码规范（置信度: 高）

Approval

• 状态: ❌ 失败
• 错误类型: 代码规范
• 置信度: 高
• 根因摘要: PR新增logger.info触发日志修改审批规则，需指定RD审批
• 分析器: 通用分析(fallback)

根因详情:
check_approval.sh 脚本检测到 PR diff 中存在日志行为修改（新增 logger.info(f"The current platform[sm{get_sm_version()}] can't import Flash Attention V4.")），触发了规则：修改 .info/.debug/.error/log_request 日志行为需要 FastDeploy RD（xyxinyang 或 zyyzghb）之一的 Review Approval。当前未满足此审批条件，脚本以 exit code 6 退出。

修复建议:

1. 请 xyxinyang (zhouchong) 或 zyyzghb (zhangyongyue) 在此 PR 上进行 Approve 审批

修复建议摘要: 请 xyxinyang 或 zyyzghb Approve 此 PR

关联变更: PR 新增了 logger.info(...) 日志语句（涉及日志行为修改）
链接: 查看日志

[codecov-commenter]

Codecov Report

❌ Patch coverage is 0% with 7 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@dad5a43). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...el_executor/layers/attention/flash_attn_backend.py	0.00%	7 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #7819   +/-   ##
==========================================
  Coverage           ?   63.33%           
==========================================
  Files              ?      462           
  Lines              ?    64371           
  Branches           ?     9872           
==========================================
  Hits               ?    40769           
  Misses             ?    20835           
  Partials           ?     2767           

Flag	Coverage Δ
GPU	72.44% <0.00%> (?)
XPU	7.12% <0.00%> (?)

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.

[zoooo0820]

LGTM

[PaddlePaddle-bot]

🤖 Paddle-CI-Agent | ci_status_monitor | 2026-05-18 10:27:43

CI报告基于以下代码生成（30分钟更新一次）:

• PR commit: 3f9ddc9
• Merge base: 12c6ae0 (branch: develop)
• 查看完整 Diff
• CI 详情

---

1 任务总览

⏳ Required 任务进行中：5 个运行中，1 个等待中，暂无 Required 失败。

总执行（rerun次数）	总任务	✅ 通过	❌ 失败	⏳ 运行中	⏸️ 等待中	跳过
37(0)	37	27	1	7	2	0

2 任务状态汇总

2.1 Required任务 : 3/9 通过

必选任务阻塞合并，失败需优先处理。

状态	任务	耗时	根因	修复建议	日志	重跑
⏳	Extracted partial CE model tasks to run in CI. / run_ce_cases	-	运行中	-	Job	-
⏳	Run Base Tests / base_tests	-	运行中	-	Job	-
⏳	Run FastDeploy Unit Tests and Coverage / run_tests_with_coverage	-	运行中	-	Job	-
⏳	xpu_4cards_case_test / run_xpu_4cards_cases	-	运行中	-	Job	-
⏳	xpu_8cards_case_test / run_xpu_8cards_cases	-	运行中	-	Job	-
⏸️	Run Four Cards Tests / run_4_cards_tests	-	等待中	-	-	-
✅	其余 3 个必选任务通过	-	-	-	-	-

2.2 可选任务 — 24/28 通过

可选任务不阻塞合并，失败仅供参考。

状态	任务	耗时	日志	重跑
❌	Trigger Jenkins for PR	10m44s	Job	-
⏳	Run iluvatar Tests / run_iluvatar_cases	-	Job	-
⏳	xpu_unit_test / run_xpu_unit_test	-	Job	-
⏸️	CI_HPU	-	-	-
✅	其余 24 个可选任务通过	-	-	-

---

3 失败详情（仅 required）

无 required 失败任务。

[PaddlePaddle-bot]

🤖 Paddle-CI-Agent | pr_review | 2026-05-19 22:59:58

📋 Review 摘要

PR 概述：将 flash mask 从独立包迁移到 paddlefleet.ops 下，并移除 requirements.txt 中的旧版本锁定依赖。
变更范围：fastdeploy/model_executor/layers/attention/flash_attn_backend.py、requirements.txt
影响面 Tag：[OP]

问题

级别	文件	概述
🔴 Bug	flash_attn_backend.py:102	fa4 变量在内层异常分支未赋值，外层继续使用导致 NameError 崩溃

📝 PR 规范检查

标题格式合规，PR 描述所有必填 section 均已填写，Checklist 已勾选，规范符合要求。✓

总体评价

PR 意图清晰，依赖迁移合理。但新增的嵌套异常处理逻辑存在明确的变量作用域 Bug——当 flash_mask 不可用时 fa4 未赋值，外层代码继续访问会抛出 NameError 导致进程崩溃，需在合入前修复。

[PaddlePaddle-bot]

🔴 Bug fa4 变量在异常分支未被赋值，但外层代码仍访问它

当 is_flash_mask_available() 返回 False 或 ImportError/ModuleNotFoundError 被内层 except 捕获时，fa4 变量并未赋值。此时外层 try 块第 102 行 flashmask_attention_v4 = fa4 会抛出 NameError: name 'fa4' is not defined，而外层 except ImportError 无法捕获 NameError，导致程序崩溃，等同于 flash_mask 完全不可用时的行为比原来更糟。

建议修复方式：

try:
    paddle.enable_compat(scope={"cutlass"})
    fa4 = None
    try:
        from paddlefleet.ops import is_flash_mask_available
        if is_flash_mask_available():
            from paddlefleet.ops.flash_mask.cute.interface import (
                flashmask_attention as fa4,
            )
        else:
            raise ModuleNotFoundError("flash_mask not available.")
    except (ImportError, ModuleNotFoundError):
        logger.info(f"The current platform[sm{get_sm_version()}] can't import Flash Attention V4.")

    if fa4 is not None:
        global flashmask_attention_v4
        flashmask_attention_v4 = fa4
        FLASH_ATTN_VERSION = 4
        logger.info("The current platform supports Flash Attention V4.")
except ImportError:
    logger.info(f"The current platform[sm{get_sm_version()}] can't import Flash Attention V4.")

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ BingooYang
❌ root

---

root seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}