[MM][Perf] Replace VisionPatchEmbed with that in vllm for better performance

[shen-shanshan]

What this PR does / why we need it?

Replace VisionPatchEmbed with that in vllm for better performance.

• TTFT (ms): has been reduced 31.81%.
• TPOT (ms): has been reduced 20.89%.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

Run:

cd /workspace/vllm-ascend
bash benchmarks/scripts/run-performance-benchmarks.sh

Before this PR:

============ Serving Benchmark Result ============
Successful requests:                     200       
Failed requests:                         0         
Request rate configured (RPS):           16.00     
Benchmark duration (s):                  45.35     
Total input tokens:                      20026     
Total generated tokens:                  20430     
Request throughput (req/s):              4.41      
Output token throughput (tok/s):         450.48    
Peak output token throughput (tok/s):    2055.00   
Peak concurrent requests:                194.00    
Total Token throughput (tok/s):          892.06    
---------------Time to First Token----------------
Mean TTFT (ms):                          11300.73  
Median TTFT (ms):                        11307.59  
P99 TTFT (ms):                           23844.70  
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          243.95    
Median TPOT (ms):                        235.41    
P99 TPOT (ms):                           454.32    
---------------Inter-token Latency----------------
Mean ITL (ms):                           219.73    
Median ITL (ms):                         79.90     
P99 ITL (ms):                            666.86    
==================================================

After this PR:

============ Serving Benchmark Result ============
Successful requests:                     200       
Failed requests:                         0         
Request rate configured (RPS):           16.00     
Benchmark duration (s):                  36.47     
Total input tokens:                      20026     
Total generated tokens:                  21020     
Request throughput (req/s):              5.48      
Output token throughput (tok/s):         576.31    
Peak output token throughput (tok/s):    2275.00   
Peak concurrent requests:                194.00    
Total Token throughput (tok/s):          1125.37   
---------------Time to First Token----------------
Mean TTFT (ms):                          7706.34   
Median TTFT (ms):                        7604.02   
P99 TTFT (ms):                           16479.20  
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          192.98    
Median TPOT (ms):                        189.42    
P99 TPOT (ms):                           352.45    
---------------Inter-token Latency----------------
Mean ITL (ms):                           171.75    
Median ITL (ms):                         73.73     
P99 ITL (ms):                            569.89    
==================================================

• vLLM version: v0.11.0
• vLLM main: vllm-project/vllm@2918c1b

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request improves performance by replacing the custom AscendQwen2_5_VisionPatchEmbed with the standard vLLM implementation. The changes, including the removal of the custom class and related weight conversion logic, are consistent and well-justified by the significant performance gains shown in the benchmarks.

However, there is a critical issue: the unit tests for the removed AscendQwen2_5_VisionPatchEmbed in tests/ut/models/test_qwen2_5_vl.py have not been deleted. This will cause an ImportError and break the build. Please remove the obsolete TestAscendQwen2_5_VisionPatchEmbed class and its import from the test file.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[shen-shanshan]

These changes have been merge into #4349.