init v0.23.0

Signed-off-by: Sun Ruoxi <sunruoxi@4paradigm.com>

tests/e2e/vllm_interface/singlecard/test_sampler.py

@@ -25,12 +25,9 @@ def test_models_topk() -> None:
     example_prompts = [
         "The capital of France is",
     ]
-    sampling_params = SamplingParams(max_tokens=10,
-                                     temperature=0.0,
-                                     top_k=10,
-                                     top_p=0.9)
+    sampling_params = SamplingParams(max_tokens=10, temperature=0.0, top_k=10, top_p=0.9)
 
-    with VllmRunner("Qwen/Qwen3-0.6B",
-                    max_model_len=4096,
-                    gpu_memory_utilization=0.7) as runner:
+    with VllmRunner(
+        "Qwen/Qwen3-0.6B", max_model_len=4096, cudagraph_capture_sizes=[1, 2, 4, 8], gpu_memory_utilization=0.7
+    ) as runner:
         runner.generate(example_prompts, sampling_params)