[featrue]Large EP support enable_micro_batch config in vllm_config and… · Pull Request !602 · MindSpore/vllm-mindspore

[featrue]Large EP support enable_micro_batch config in vllm_config and forward phase: prefill_micro_batch
1. update vllm_mindspore/model_executor/models/mf_models/deepseek_v3.py
2. update vllm_mindspore/model_executor/models/mf_models/mf_model_base.py
3. update vllm_mindspore/model_executor/models/model_base.py
---
<!--  Thanks for sending a pull request!  Here are some tips for you:

1) If this is your first time, please read our contributor guidelines: https://gitee.com/mindspore/mindspore/blob/master/CONTRIBUTING.md

2) If you want to contribute your code but don't know who will review and merge, please add label `mindspore-assistant` to the pull request, we will find and do it as soon as possible.
-->

**What type of PR is this?**
/kind feature

**What does this PR do / why do we need it**:
1. Bypass  **additional_config.enbale_micro_batch**  to enable large EP computation and communication overlap.
2. When enable micro batch and input batch_size > 1, use graph phase " **prefill_micro_batch** ".
3. Test record:
comman: vllm-mindspore serve --model="/data3/ckpt/dsr1-w8a8/" --trust_remote_code --max-num-seqs=256 --max_model_len=4096 --max-num-batched-tokens=256 --block-size=128 --gpu-memory-utilization=0.93 --tensor-parallel-size 1 --data-parallel-size 16 --data-parallel-size-local 16 --enable-expert-parallel  **--additional-config '{"enable_micro_batch": 1}'**  >log 2>&1 &
4. Enable micro batch profiling:
![输入图片说明](https://foruda.gitee.com/images/1749483023792056035/7aeeb9d0_12317514.png "屏幕截图")

**Which issue(s) this PR fixes**:

Fixes #SR.IR20250325004551.004

**Code review checklist [【代码检视checklist说明】](https://gitee.com/mindspore/community/blob/master/security/code_review_checklist_mechanism.md)**:

+ - [ ] 是否进行返回值校验 (禁止使用void屏蔽安全函数、自研函数返回值，C++标准库函数确认无问题可以屏蔽)
+ - [ ] 是否遵守 ***SOLID原则 / 迪米特法则***
+ - [ ] 是否具备UT测试用例看护 && 测试用例为有效用例 (若新特性无测试用例看护请说明原因)
+ - [ ] 是否为对外接口变更
+ - [ ] 是否涉及官网文档修改

类型	指派人员	状态
审查	TronZhang	已完成 (0/0人)
测试	TronZhang	已完成 (0/0人)

MindSpore/vllm-mindspore .gitee-modal { width: 500px !important; }

搜索帮助

MindSpore/vllm-mindspore