[release/0.3][Bug fixes] support csa_sparse_attn_bwd_cudnn for h-card

[risemeup1111]

PR Category

Environment Adaptation

PR Types

Bug fixes

Description

support csa_sparse_attn_bwd_cudnn for h-card

是否引起精度变化

否

Cherry-pick of #1472 to release/0.3.

Merged in dev: #1472

[Paddle-CI-Bot]

PaddleFleet Log Analysis

Run #29893112015 · Attempt 1

日志分析报告

流水线名称	问题标签	修复建议	日志片段
Build Fleet whl / Check docker image and run container	Docker Registry 网络超时	与本PR无关，CCR 镜像仓库认证服务连接超时，CI 维护人员检查 runner 网络/代理出口至 ccr-auth.bj.baidubce.com 的连通性，重新触发即可	报错代码

失败的测试case:

无测试 case 失败。失败发生在构建阶段（Check docker image and run container），
docker pull 时 CCR 认证请求超时，容器从未启动，后续所有 job（uv build whl、
ops whl、单测、集成测试）均因此跳过（skipped）。

根本原因分析:
docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda12.9-cudnn9.9-trt10.5-gcc11 时，向 ccr-auth.bj.baidubce.com 请求 token 超时（Client.Timeout exceeded while awaiting headers），导致镜像无法拉取，与 PR 1486 代码改动无关。

修复建议:

1. CI 维护人员确认 runner（paddle-2）出口代理至 ccr-auth.bj.baidubce.com 的 HTTPS 连通性；
2. 直接在 runner 上执行 curl -v https://ccr-auth.bj.baidubce.com 验证是否超时；
3. 网络恢复后重新触发该 workflow 即可，无需修改 PR 代码。

---

🔍 准确性记录：请点击评论底部 😊 图标，选择 👍（准确）或 👎（有误），将自动记录到 CI 监控系统

_{🔄 每次 Re-run 后自动更新}

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (release/0.3@f5b681d). Learn more about missing BASE report.

Additional details and impacted files

@@               Coverage Diff               @@
##             release/0.3     #1486   +/-   ##
===============================================
  Coverage               ?   100.00%           
===============================================
  Files                  ?         1           
  Lines                  ?        52           
  Branches               ?         6           
===============================================
  Hits                   ?        52           
  Misses                 ?         0           
  Partials               ?         0           

Flag	Coverage Δ
coverage_combine	100.00% <100.00%> (?)

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

Files with missing lines	Coverage Δ
...efleet/cudnn_ops/attn/csa_sparse_attn_bwd_cudnn.py	100.00% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.