[VLM] Support Piecewise CUDA Graph for Qwen2.5-VL (#13055)

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
Co-authored-by: Yuhao Yang <yhyang201@gmail.com>

test/srt/run_suite.py

@@ -94,7 +94,7 @@ suites = {
         TestFile("test_original_logprobs.py", 41),
         TestFile("test_page_size.py", 60),
         TestFile("test_penalty.py", 82),
-        TestFile("test_piecewise_cuda_graph.py", 600),
+        TestFile("test_piecewise_cuda_graph.py", 750),
         TestFile("test_priority_scheduling.py", 130),
         TestFile("test_pytorch_sampling_backend.py", 66),
         TestFile("test_radix_attention.py", 105),
@@ -287,6 +287,7 @@ suites = {
         TestFile("nightly/test_text_models_perf.py"),
         TestFile("nightly/test_vlms_mmmu_eval.py"),
         TestFile("nightly/test_vlms_perf.py"),
+        TestFile("nightly/test_vlms_piecewise_cuda_graph.py"),
         TestFile("test_quick_allreduce.py"),
         TestFile("test_sagemaker_server.py"),
         TestFile("test_schedule_policy.py"),