fp8 triton moe config.

[xuanyuanminzheng]

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.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 20.00000% with 4 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@9431c4f). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...el_executor/layers/moe/fused_moe_triton_backend.py	20.00%	3 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #8036   +/-   ##
==========================================
  Coverage           ?   67.66%           
==========================================
  Files              ?      470           
  Lines              ?    66115           
  Branches           ?    10189           
==========================================
  Hits               ?    44740           
  Misses             ?    18523           
  Partials           ?     2852           

Flag	Coverage Δ
GPU	77.76% <20.00%> (?)
XPU	7.01% <0.00%> (?)

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

☔ View full report in Codecov by Harness.
📢 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.

[PaddlePaddle-bot]

🤖 Paddle-CI-Agent | pr_review | 2026-06-11 00:03:39

📋 Review 摘要

PR 概述：调整 FP8 Triton MoE 在小 token 场景下的 tiling 配置。
变更范围：fastdeploy/model_executor/layers/moe/fused_moe_triton_backend.py
影响面 Tag：[OP] [Quantization] [Optimization]

问题

未发现阻塞性问题。PR 规范问题在下面章节报,不要在这里重复

📝 PR 规范检查

标题缺少官方 Tag，PR 描述仍是模板占位内容，未填写 Motivation、Modifications、Usage or Command、Accuracy Tests 和 Checklist 状态。建议替换为以下内容。

标题建议（可直接复制）：

• [Optimization] Tune FP8 Triton MoE small-token config

PR 描述建议（点击展开，可直接复制）

## Motivation
Tune the Triton MoE FP8 kernel configuration for small-token workloads to use smaller M/N tiles and adjusted pipeline stages.

## Modifications
- In `fastdeploy/model_executor/layers/moe/fused_moe_triton_backend.py`, refine the `Wfp8Afp8MoEMethod.apply` config when `token_num <= E`:
  - `token_num <= 16`: use `BLOCK_SIZE_M=16`, `BLOCK_SIZE_N=64`, `num_stages=4`.
  - `token_num <= 32`: use `BLOCK_SIZE_M=32`, `BLOCK_SIZE_N=64`, `num_stages=3`.
  - `token_num > 32` and `token_num <= E`: keep `BLOCK_SIZE_M=64`, `BLOCK_SIZE_N=128`, and set `num_stages=3`.

## Usage or Command
N/A

## Accuracy Tests
N/A

## Checklist

- [x] 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.

总体评价

本轮按风险优先审查了 FP8 Triton MoE 小 token 配置分支，以及该配置传入 preprocess 和两次 fused MoE GEMM kernel 的调用关系；未确认到会改变路由、shape、scale 或写回语义的阻塞性问题。建议补齐 PR 标题和描述中的规范信息，尤其是性能/精度验证结果或未提供的原因。

[PaddlePaddle-bot]

🤖 Paddle-CI-Agent | ci_status_monitor | 2026-06-14 04:05:18 UTC+08:00

CI报告基于以下代码生成（30分钟更新一次）:
PR commit: c083797 | Merge base: 9431c4f (branch: develop)

---

1 Required任务 : 9/10 通过

总执行（rerun次数）	总任务	✅ 通过	❌ 失败	⏳ 运行中	⏸️ 等待中	跳过
41(0)	41	37	4	0	0	0

任务	错误类型	置信度	日志
Run FastDeploy Unit Tests and Coverage / run_tests_with_coverage	PR问题：diff coverage 低于阈值	高	Job

2 失败详情

🔴 Run FastDeploy Unit Tests and Coverage / run_tests_with_coverage — PR问题（置信度: 高）

错误类型: PR问题 | 置信度: 高
分析器: 通用分析(fallback)
失败用例: 覆盖率门禁

用例	错误摘要
diff coverage gate	fused_moe_triton_backend.py 新增分支覆盖率 40%，低于 80% 阈值

关键日志:

All tests passed
Failure. Coverage is below 80%.
fastdeploy/model_executor/layers/moe/fused_moe_triton_backend.py (40.0%): Missing lines 759-760,769
Total:   5 lines
Missing: 3 lines
Coverage: 40%
COVERAGE_EXIT_CODE: 9

• 根因摘要: 新增 token 分桶分支缺少 diff coverage
  PR 只修改了 fastdeploy/model_executor/layers/moe/fused_moe_triton_backend.py，在 TritonMoEMethod.apply() 中新增了 token_num <= 16、token_num <= 32、else 三段配置分支。CI 单测已通过，但 diff coverage 只覆盖了新增/变更 5 行中的 2 行，缺失第 759-760 行的 16 < token_num <= 32 分支和第 769 行的 token_num > 32 分支，因此覆盖率门禁退出码为 9。

修复建议:

1. 在 tests/layers/test_fused_moe_triton_backend.py 的 TestTritonMoEMethod 的 apply 测试区补充轻量单测，构造 17 <= token_num <= 32 并断言 kernel 参数包含 BLOCK_SIZE_M=32、num_stages=3。
2. 再补一个 33 <= token_num <= num_local_experts 的用例，断言 kernel 参数包含 BLOCK_SIZE_M=64、BLOCK_SIZE_N=128、num_stages=3，覆盖 else 分支入口。
3. 复用现有 fake_ops、_patch_bf16_kernel() 和 _make_layer(num_experts=...)，避免触发真实 Triton kernel；本次失败不需要修改 CI 环境。

关联变更: fastdeploy/model_executor/layers/moe/fused_moe_triton_backend.py:747 附近新增配置分支；缺覆盖行 759-760,769。