[Refactor] native model use set_graph_name api · Pull Request !521 · MindSpore/vllm-mindspore

1. refactor native model jit
---
<!--  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 refactor

**What does this PR do / why do we need it**:
重构Native模型的jit写法：
1. 显示调用两次jit编译pa/fa图，后续选择不同图执行，便于开发者理解；
2. 使用_set_jit_graph_name接口使能编图，Cell.phase字段当前仅作为性能优化功能，后续使用优化方案代替；
3. 使用全局属性代替is_prefill入参，减少模型入参传递；

**V1自验结果** ：
```
VLLM_USE_V1=1 vllm-mindspore serve /home/ckpt/qwen2.5-7b-hf --port=8812  --tensor_parallel_size=2
```
单请求带后处理：
```
curl http://0.0.0.0:8812/v1/completions -H "Content-Type: application/json" -d '{"model": "/home/ckpt/qwen2.5-7b-hf", "prompt": "You are a helpful assistant.<｜User｜>将文本分类为中性、负面或正面。 \n文本：我认为这次假期还可以。 \n情感：<｜Assistant｜><think>\n", "max_tokens": 256, "temperature": 0.7, "top_p": 0.8, "top_k": 20, "repetition_penalty":1.05}'
```
{"id":"cmpl-1f4718defab34a58a41b066c28c8d1c4","object":"text_completion","created":1750497947,"model":"/home/ckpt/qwen2.5-7b-hf","choices":[{"index":0,"text":"这句话表达了一种中性的态度，既不是特别积极也不是特别消极，只是简单地表示对假期的感受是“还可以”，没有明显的积极或消极情绪。\n\n分类：中性</think>","logprobs":null,"finish_reason":"stop","stop_reason":151643,"prompt_logprobs":null}],"usage":{"prompt_tokens":38,"total_tokens":80,"completion_tokens":42,"prompt_tokens_details":null}}

性能验证：
```
python benchmark_parallel.py --backend openai --host 90.90.94.238 --port 8812  --tokenizer /home/ckpt/qwen2.5-7b-hf  --epochs 1 --parallel-num 192 --prompt-tokens 256 --output-tokens 256
```
所有请求耗时: 15.8277 s
请求吞吐: 12.1306 requests/s
输出tokens总吞吐: 3105.4431 tokens/s
首tokens时延TP90: 698.8264 ms
首tokens时延TP99: 703.1921 ms
最大首tokens时延: 706.0198 ms
平均首tokens时延: 568.8561 ms
增量时延TP90: 62.2489 ms
增量时延TP99: 83.1290 ms
最大增量时延: 105.8 ms
平均增量时延: 59.3 ms

**V0自验结果** ：
```
VLLM_USE_V1=0 vllm-mindspore serve /home/ckpt/qwen2.5-7b-hf --port=8812  --tensor_parallel_size=2
```
单请求带后处理：
```
curl http://0.0.0.0:8812/v1/completions -H "Content-Type: application/json" -d '{"model": "/home/ckpt/qwen2.5-7b-hf", "prompt": "You are a helpful assistant.<｜User｜>将文本分类为中性、负面或正面。 \n文本：我认为这次假期还可以。 \n情感：<｜Assistant｜><think>\n", "max_tokens": 256, "temperature": 0.7, "top_p": 0.8, "top_k": 20, "repetition_penalty":1.05}'
```
{"id":"cmpl-a1c3c47241734c7abd99e232f829e6ea","object":"text_completion","created":1750498621,"model":"/home/ckpt/qwen2.5-7b-hf","choices":[{"index":0,"text":"这句话表达了一种中性的观点，既没有明显的积极情绪也没有明显的消极情绪。\n</think>中性<｜Assistant｜>","logprobs":null,"finish_reason":"stop","stop_reason":151643,"prompt_logprobs":null}],"usage":{"prompt_tokens":38,"total_tokens":67,"completion_tokens":29,"prompt_tokens_details":null}}

性能验证：

```
python benchmark_parallel.py --backend openai --host 90.90.94.238 --port 8812  --tokenizer /home/ckpt/qwen2.5-7b-hf  --epochs 1 --parallel-num 192 --prompt-tokens 256 --output-tokens 256
```
所有请求耗时: 21.9501 s
请求吞吐: 8.7471 requests/s
输出tokens总吞吐: 2239.2621 tokens/s
首tokens时延TP90: 3434.7362 ms
首tokens时延TP99: 3440.9919 ms
最大首tokens时延: 3441.7510 ms
平均首tokens时延: 2661.1855 ms
增量时延TP90: 73.4981 ms
增量时延TP99: 110.2794 ms
最大增量时延: 842.5 ms
平均增量时延: 72.5 ms

**动态图验证** ：
```
VLLM_USE_V1=1 vllm-mindspore serve /home/ckpt/qwen2.5-7b-hf --port=8812  --tensor_parallel_size=2  --enforce-eager
```
{"id":"cmpl-97f0d556b21f44cdba9881d162b74827","object":"text_completion","created":1750502708,"model":"/home/ckpt/qwen2.5-7b-hf","choices":[{"index":0,"text":"这句话表达了一种中立的态度，既没有明显的积极情绪也没有明显的消极情绪。\n\n分类结果：中性\n</think>","logprobs":null,"finish_reason":"stop","stop_reason":151643,"prompt_logprobs":null}],"usage":{"prompt_tokens":38,"total_tokens":66,"completion_tokens":28,"prompt_tokens_details":null}}

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

Fixes #ICB1V2

**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