[Feature] support nvfp4 tbo

[lizexu123]

Motivation

PR 概述：为 NVFP4 MoE 添加 TBO (Two Batch Overlap) 支持

💡 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.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ lizexu123
❌ root

---

root seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[paddle-bot]

Thanks for your contribution!

[codecov-commenter]

Codecov Report

❌ Patch coverage is 7.50000% with 37 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@48d2bbe). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...deploy/model_executor/layers/quantization/nvfp4.py	7.50%	37 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #7259   +/-   ##
==========================================
  Coverage           ?   73.84%           
==========================================
  Files              ?      383           
  Lines              ?    53548           
  Branches           ?     8390           
==========================================
  Hits               ?    39543           
  Misses             ?    11238           
  Partials           ?     2767           

Flag	Coverage Δ
GPU	73.84% <7.50%> (?)

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.

[PaddlePaddle-bot]

📋 Review 摘要

PR 概述：为 NVFP4 MoE 添加 TBO (Thread Batching Optimization) 支持

变更范围：model_executor/layers/quantization/nvfp4.py

影响面 Tag：[Feature] [Quantization]

📝 PR 规范检查

✅ PR 标题 [Feature] support nvfp4 tbo 符合规范
✅ PR 描述填写了 Motivation 字段
⚠️ Checklist 未完成（未勾选任何选项）

建议补全 Checklist：

• Add at least a tag in the PR title. [Feature]
• Format your code, run pre-commit before commit.
• Add unit tests. Please write the reason in this PR if no unit tests.

问题

级别	文件	概述
🟡 建议	nvfp4.py:708	global_values 全局字典只写入未读取，疑似调试代码遗留
🟡 建议	nvfp4.py:730	大量 tensor 被存储到全局字典，可能导致内存泄漏

总体评价

PR 实现了 NVFP4 MoE 的 TBO 支持，整体逻辑正确。与 fused_moe_blackwell_backend.py 和 fused_moe_deepgemm_backend.py 的实现模式保持一致。存在一处遗留的调试代码，建议删除以避免潜在内存泄漏问题。

[PaddlePaddle-bot]

🟡 建议 global_values 全局字典只写入未读取，疑似调试代码遗留。

整个代码库中没有任何地方读取 global_values 中存储的 tensor。建议删除相关代码（第 708-731 行、797 行、807 行）或添加注释说明其调试用途。

[PaddlePaddle-bot]

🟡 建议 大量 tensor 被存储到全局字典但从未读取，可能导致内存泄漏。

global_values 是模块级全局变量，存储了 x、recv_x_value、handle 等多个 tensor 对象。这些对象会被持续引用，在高并发场景下可能导致内存无法及时回收。

建议：如果这些存储确实用于调试，应在生产环境中移除；如果需要保留，考虑使用 weakref 或定期清理机制。