[Optimize] Qwen2.5-VL vision model with merged linear layers and unif…

[xiaoxiaohehe001]

Modifications

1. 合并gate_proj和up_proj为up_gate_proj
2. 使用统一的RMSNorm层

before

============ Serving Benchmark Result ============
Successful requests:                     2350      
Benchmark duration (s):                  1011.96   
Total input tokens:                      6591406   
Total generated tokens:                  3525000   
Request throughput (req/s):              2.322     
Output token throughput (tok/s):         3483.35   
Total Token throughput (tok/s):          9996.89   
-------------------解码速度(tok/s)--------------------
Mean Decode:                             32.78     
Median Decode:                           32.35     
P99 Decode:                              67.39     
---------------Time to First Token----------------
Mean TTFT (ms):                          8115.94   
Median TTFT (ms):                        7694.97   
P99 TTFT (ms):                           29780.83  
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          30.93     
Median TPOT (ms):                        31.06     
P99 TPOT (ms):                           36.83     
---------------Inter-token Latency----------------
Mean ITL (ms):                           30.90     
Median ITL (ms):                         17.51     
P99 ITL (ms):                            610.61    
----------------End-to-end Latency----------------
Mean E2EL (ms):                          54472.78  
Median E2EL (ms):                        54733.45  
P99 E2EL (ms):                           67343.39  
-------------Infer End-to-end Latency-------------
Mean S_E2EL (ms):                        47336.72  
Median S_E2EL (ms):                      47152.32  
P99 S_E2EL (ms):                         56024.31  
==================================================

after

============ Serving Benchmark Result ============
Successful requests:                     2350      
Benchmark duration (s):                  948.40    
Total input tokens:                      6591406   
Total generated tokens:                  3525000   
Request throughput (req/s):              2.478     
Output token throughput (tok/s):         3716.79   
Total Token throughput (tok/s):          10666.82  
-------------------解码速度(tok/s)--------------------
Mean Decode:                             34.02     
Median Decode:                           33.08     
P99 Decode:                              60.80     
---------------Time to First Token----------------
Mean TTFT (ms):                          6422.78   
Median TTFT (ms):                        5483.12   
P99 TTFT (ms):                           26078.27  
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          29.78     
Median TPOT (ms):                        30.20     
P99 TPOT (ms):                           34.54     
---------------Inter-token Latency----------------
Mean ITL (ms):                           29.76     
Median ITL (ms):                         17.39     
P99 ITL (ms):                            542.75    
----------------End-to-end Latency----------------
Mean E2EL (ms):                          51059.96  
Median E2EL (ms):                        51309.52  
P99 E2EL (ms):                           70451.84  
-------------Infer End-to-end Latency-------------
Mean S_E2EL (ms):                        45460.27  
Median S_E2EL (ms):                      45970.20  
P99 S_E2EL (ms):                         51969.68    
==================================================

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!

[codecov-commenter]

Codecov Report

❌ Patch coverage is 95.45455% with 1 line in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@273e79a). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...del_executor/models/qwen2_5_vl/dfnrope/modeling.py	95.23%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #6037   +/-   ##
==========================================
  Coverage           ?   67.29%           
==========================================
  Files              ?      348           
  Lines              ?    44765           
  Branches           ?     6891           
==========================================
  Hits               ?    30123           
  Misses             ?    12419           
  Partials           ?     2223           

Flag	Coverage Δ
GPU	67.29% <95.45%> (?)

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.