[Cherry-Pick][BugFix] support deepgemm without bias input(#7559)

[EmmonsCurse]

Cherry-pick of #7559 (authored by @zoooo0820) to release/2.6.

devPR:#7559

---

Motivation

适配deepgemm缺少bias参数时的处理

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.

[paddle-bot]

Thanks for your contribution!

[PaddlePaddle-bot]

🤖 AI Code Review | 2026-04-22 19:51:03

📋 Review 摘要

PR 概述：适配 DeepGEMM FP8 GEMM 算子在不支持 bias 关键字参数时的兼容处理，通过模块加载时运行时探测自动选择调用路径。
变更范围：model_executor/layers/quantization/block_wise_fp8.py
影响面 Tag：OP Quantization

---

📝 PR 规范检查

Modifications 节未填写，建议补充说明本次变更的具体修改内容，供可直接复制的示例如下：

Modifications（可直接复制）：

• 在模块加载时通过 inspect.signature 探测 fp8_gemm_nt 是否支持 bias 关键字参数，结果缓存至全局变量 _fp8_gemm_nt_has_bias_kwarg。
• deep_gemm_fp8_gemm_nt 函数中增加分支：支持 bias 时原样调用，不支持时先调用无 bias 版本再通过 paddle.add 追加 bias。

---

问题

级别	文件	概述
🟡 建议	block_wise_fp8.py:62	异常捕获注释与实现不符，误导读者
🟡 建议	block_wise_fp8.py:157	paddle.add 创建额外 Tensor，可考虑原地操作

---

总体评价

整体思路清晰，运行时 capability detection 是处理 pybind11 函数签名差异的合理方案，兼容性设计正确。建议修正误导性注释并评估 paddle.add 的显存影响后合入。

[PaddlePaddle-bot]

🟡 建议 额外 Tensor 分配

linear_out = paddle.add(linear_out, bias) 会产生一个新 Tensor，而非原地操作，在大 batch / 长序列场景下会带来额外显存峰值。建议评估是否可使用 paddle.add_(linear_out, bias) 原地加法，或与上游确认 linear_out 的生命周期是否允许原地操作。

[codecov-commenter]

Codecov Report

❌ Patch coverage is 38.46154% with 8 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (release/2.6@3d6d3a2). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...del_executor/layers/quantization/block_wise_fp8.py	38.46%	7 Missing and 1 partial ⚠️

Additional details and impacted files

@@              Coverage Diff               @@
##             release/2.6    #7565   +/-   ##
==============================================
  Coverage               ?   73.75%           
==============================================
  Files                  ?      376           
  Lines                  ?    53097           
  Branches               ?     8303           
==============================================
  Hits                   ?    39161           
  Misses                 ?    11188           
  Partials               ?     2748           

Flag	Coverage Δ
GPU	73.75% <38.46%> (?)

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.