[Cherry-Pick] CP 3 prs which is about fixing and optimizing(#7237)(#7425)(#7426)

[RichardWooSJTU]

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!

[PaddlePaddle-bot]

🤖 AI Code Review | 2026-04-16 19:56 CST

📋 Review 摘要

PR 概述：Cherry-pick 3 个 PR（#7237, #7425, #7426），包含 num_max_dispatch_tokens_per_rank 自动计算、数据并行启动优化、以及 DeepSeek V3 模型 deep_gemm 导入统一。
变更范围：config.py、engine/engine.py、model_executor/models/deepseek_v3.py
影响面 Tag：FDConfig Engine Models

📝 PR 规范检查

PR 标题缺少功能性 Tag（仅有 [Cherry-Pick]），且 Motivation / Modifications 章节未填写。

标题建议（可直接复制）：

• [Cherry-Pick][BugFix][Optimization] CP 3 prs which is about fixing and optimizing(#7237)(#7425)(#7426)

描述建议：请补充 Motivation（说明这 3 个原始 PR 分别解决什么问题）和 Modifications（概述三处变更的具体内容）。

问题

级别	文件	概述
🟡 建议	config.py:2166	getattr truthiness 检查可改用 is not None 更清晰

总体评价

三处变更逻辑正确：config 自动计算 dispatch tokens 确保一致性、engine 并行启动优化减少等待时间、deepseek_v3 统一使用 fp8_utils 集中管理 deep_gemm 导入。仅有一处代码风格建议，无阻塞性问题。

[PaddlePaddle-bot]

🟡 建议 getattr(..., None) 依赖 truthiness 判断属性是否存在，当值恰好为 0 或其他 falsy 值时会被跳过。

虽然 num_max_dispatch_tokens_per_rank 默认值为 128，实际不太可能为 0，但使用 is not None 能更精确地表达"属性存在即覆盖"的意图，避免潜在的边界情况。

建议修改为：

if (
    getattr(self.model_config, "num_max_dispatch_tokens_per_rank", None) is not None
    and self.model_config.num_max_dispatch_tokens_per_rank != auto_dispatch_tokens
):