[Refactor] Cuda Graph Runner/Backend Refactor (#23906)

Co-authored-by: BBuf <1182563586@qq.com>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>

test/registered/quant/test_fp8_gemm_sm120.py

@@ -33,7 +33,7 @@ class FP8GemmSM120Base:
             "--trust-remote-code",
             "--fp8-gemm-backend",
             cls.backend,
-            "--disable-piecewise-cuda-graph",
+            "--cuda-graph-backend-prefill=disabled",
         ]
         if cls.quantization:
             other_args += ["--quantization", cls.quantization]

test/registered/quant/test_gguf.py

@@ -19,7 +19,9 @@ class TestGGUF(CustomTestCase):
             filename="qwen2-1_5b-instruct-q4_k_m.gguf",
         )
 
-        engine = sgl.Engine(model_path=model_path, random_seed=42, cuda_graph_max_bs=2)
+        engine = sgl.Engine(
+            model_path=model_path, random_seed=42, cuda_graph_max_bs_decode=2
+        )
         outputs = engine.generate(prompt, sampling_params)["text"]
         engine.shutdown()
 

test/registered/quant/test_nvfp4_gemm_sm120.py

@@ -32,7 +32,7 @@ class FP4GemmSM120Base:
             "modelopt_fp4",
             "--fp4-gemm-backend",
             cls.backend,
-            "--disable-piecewise-cuda-graph",
+            "--cuda-graph-backend-prefill=disabled",
         ]
         cls.process = popen_launch_server(
             cls.model,