support bf16 moe tp permute, group_gemm, unpermute

[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 40.00000% with 18 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (release/2.5@5666993). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...l_executor/layers/moe/fused_moe_cutlass_backend.py	40.00%	13 Missing and 5 partials ⚠️

Additional details and impacted files

@@              Coverage Diff               @@
##             release/2.5    #7194   +/-   ##
==============================================
  Coverage               ?   69.47%           
==============================================
  Files                  ?      390           
  Lines                  ?    54384           
  Branches               ?     8575           
==============================================
  Hits                   ?    37786           
  Misses                 ?    13869           
  Partials               ?     2729           

Flag	Coverage Δ
GPU	69.47% <40.00%> (?)

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.

[fastdeploy-bot]

🤖 AI Code Review | 2026-04-07 14:42 CST

📋 Review 摘要

PR 概述：为 Cutlass MoE 后端添加 bf16 数据类型的 moe_permute、group_gemm、moe_unpermute 支持，通过 FD_USE_PHI_MOE_PERMUTE 环境变量控制新代码路径
变更范围：model_executor/layers/moe/
影响面 Tag：OP Models

📝 PR 规范检查

PR 标题缺少有效的功能标签 Tag。

标题建议（可直接复制）：

• [OP] support bf16 moe tp permute, group_gemm, unpermute

描述建议：建议在 Motivation 和 Modifications 部分补充以下内容：

• Motivation：说明为何需要在 Cutlass 后端支持 bf16 MOE permute 路径
• Modifications：说明新增 deep_batch_gemm 函数和 FD_USE_PHI_MOE_PERMUTE 分支的设计

问题

级别	文件	概述
🟡 建议	fused_moe_cutlass_backend.py:59	paddlefleet_ops 可能为 None，缺少防御性检查
🟡 建议	fused_moe_cutlass_backend.py:352	permute_scale 变量未被使用

总体评价

代码逻辑正确，与 fused_moe_deepgemm_backend.py 中的实现模式保持一致。建议添加对 paddlefleet_ops 的防御性检查以提高代码健壮性。测试用例已在 tests/layers/test_deepgemm_fused_moe.py 中覆盖。

[fastdeploy-bot]

🟡 建议 paddlefleet_ops 可能为 None，缺少防御性检查

paddlefleet_ops 是通过 try_import(["paddlefleet.ops"]) 导入的，当 paddlefleet 模块不可用时会返回 None。如果用户在未安装 paddlefleet 的环境下启用 FD_USE_PHI_MOE_PERMUTE，此处会抛出 AttributeError。

建议添加防御性检查：

def deep_batch_gemm(x, y, expert_idx_per_token):
    if paddlefleet_ops is None:
        raise RuntimeError(
            "paddlefleet.ops is required for FD_USE_PHI_MOE_PERMUTE=1. "
            "Please install paddlefleet or disable this feature."
        )
    out = paddle.empty([x.shape[0], y.shape[-1]], dtype=x.dtype)
    paddlefleet_ops.deep_gemm.m_grouped_bf16_gemm_nn_contiguous(x, y, out, expert_idx_per_token)
    return out

[fastdeploy-bot]

🟡 建议 permute_scale 变量未被使用

moe_permute 返回的 permute_scale 在后续代码中未被使用。如果确实不需要，建议用 _ 替代以明确表示该变量被有意忽略：

(
    permute_input,
    permute_indices_per_token,
    topk_weights,
    _,  # permute_scale not used in bf16 path
    expert_idx_per_token,
) = paddle.nn.functional.moe_permute(...)

注意：行 423 处也存在相同情况。