[Cherry-Pick][RL]moe bf16 ep support paddle batch_gemm(#7337)

[ckl117]

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)

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!

[codecov-commenter]

Codecov Report

❌ Patch coverage is 66.66667% with 2 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (release/2.6@2ac9b89). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...l_executor/layers/moe/fused_moe_cutlass_backend.py	66.66%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@              Coverage Diff               @@
##             release/2.6    #7339   +/-   ##
==============================================
  Coverage               ?   73.84%           
==============================================
  Files                  ?      376           
  Lines                  ?    52964           
  Branches               ?     8269           
==============================================
  Hits                   ?    39112           
  Misses                 ?    11116           
  Partials               ?     2736           

Flag	Coverage Δ
GPU	73.84% <66.66%> (?)

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.

[fastdeploy-bot]

🤖 AI Code Review | 2026-04-11 19:35 CST

📋 Review 摘要

PR 概述：为 MoE BF16 EP 路径添加 paddle.batched_gemm 支持，替代原有的 compute_ffn 调用，并根据 FD_MOE_PROB_IN_ADVANCE 环境变量选择 fused_swiglu_scale 或 swiglu 激活函数。

变更范围：model_executor/layers/moe/fused_moe_cutlass_backend.py、tests/layers/test_fused_moe_cutlass_backend.py

影响面 Tag：[RL] [OP]

📝 PR 规范检查

标题问题：标题格式正确，包含 [RL] 和 [Cherry-Pick] 标签。

描述问题：Motivation 和 Modifications 章节均为空，缺少变更说明。

描述建议（可直接复制）：

## Motivation

在 MoE EP prefill 路径中，当 `FD_USE_PHI_MOE_PERMUTE=True` 且使用 w16a16 量化时，使用 `paddle.batched_gemm` 替代原有的 `compute_ffn` 调用，以提高性能。同时支持 `FD_MOE_PROB_IN_ADVANCE` 环境变量控制是否在激活函数前进行概率加权。

## Modifications

1. 在 `apply_ep_prefill` 方法的 `moe_permute/moe_unpermute` 路径中：
   - 使用两次 `paddle.incubate.nn.functional.batched_gemm` 替代 `self.compute_ffn` 调用
   - 根据 `FD_MOE_PROB_IN_ADVANCE` 环境变量选择：
     - True: 使用 `paddlefleet_ops.fused_swiglu_scale` (融合激活和缩放)
     - False: 使用 `paddle.incubate.nn.functional.swiglu` (仅激活)
   - 调整 `moe_unpermute` 的 `using_weighted_combine` 参数逻辑

2. 测试文件：
   - 添加 `align` 辅助函数用于对齐 token 数
   - 修改 `RealMoELayer` 的权重形状定义以适配 batched_gemm
   - 在 stub dispatch 中添加对 `expert_alignment` 的支持

问题

级别	文件	概述
🔴 Bug	fused_moe_cutlass_backend.py:176	paddlefleet_ops 可能为 None，直接访问属性会导致 AttributeError

总体评价

整体变更思路合理，使用 batched_gemm 替代原有 FFN 计算路径。但存在一个运行时错误风险：当 paddlefleet.ops 不可用时，直接调用 paddlefleet_ops.fused_swiglu_scale 会抛出异常，需要添加可用性检查。

[fastdeploy-bot]

🔴 Bug 当 paddlefleet.ops 不可用时，此处会抛出 AttributeError。paddlefleet_ops 通过 try_import 导入，失败时返回 None，直接访问属性会导致运行时错误。

建议修复方式：

if fastdeploy.envs.FD_MOE_PROB_IN_ADVANCE:
    if paddlefleet_ops is not None and hasattr(paddlefleet_ops, "fused_swiglu_scale"):
        out = paddlefleet_ops.fused_swiglu_scale(out, dst_weights)
    else:
        out = paddle.incubate.nn.functional.swiglu(out)
else:
    out = paddle.incubate.nn.functional.swiglu(out)