[None][fix] fix CUDA graph config for test_llm_api_pytorch.py. (#6826)

Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>

Author: yuxianq

tests/integration/defs/accuracy/test_llm_api_pytorch.py

@@ -561,7 +561,8 @@ def test_fp8(self, cuda_graph, tp_size, pp_size, ep_size):
                 max_seq_len=8192,
                 pipeline_parallel_size=pp_size,
                 moe_expert_parallel_size=ep_size,
-                use_cuda_graph=cuda_graph) as llm:
+                cuda_graph_config=CudaGraphConfig()
+                if cuda_graph else None) as llm:
             assert llm.args.quant_config.quant_algo == QuantAlgo.FP8
             assert llm.args.quant_config.kv_cache_quant_algo == QuantAlgo.FP8
             task = MMLU(self.MODEL_NAME)
@@ -584,7 +585,8 @@ def test_fp8_chunked_prefill(self, cuda_graph, tp_size, pp_size, ep_size):
                 moe_expert_parallel_size=ep_size,
                 enable_chunked_prefill=True,
                 max_num_tokens=256,
-                use_cuda_graph=cuda_graph) as llm:
+                cuda_graph_config=CudaGraphConfig()
+                if cuda_graph else None) as llm:
             assert llm.args.quant_config.quant_algo == QuantAlgo.FP8
             assert llm.args.quant_config.kv_cache_quant_algo == QuantAlgo.FP8
             task = MMLU(self.MODEL_NAME)
@@ -704,7 +706,8 @@ def test_fp8_chunked_prefill(self, cuda_graph, tp_size, pp_size, ep_size):
                 moe_expert_parallel_size=ep_size,
                 enable_chunked_prefill=True,
                 max_num_tokens=256,
-                use_cuda_graph=cuda_graph) as llm:
+                cuda_graph_config=CudaGraphConfig()
+                if cuda_graph else None) as llm:
             assert llm.args.quant_config.quant_algo == QuantAlgo.FP8
             assert llm.args.quant_config.kv_cache_quant_algo == QuantAlgo.FP8
             task = MMLU(self.MODEL_NAME)
@@ -726,7 +729,8 @@ def test_fp4_chunked_prefill(self, cuda_graph, tp_size, pp_size, ep_size):
                 max_seq_len=22000,
                 enable_chunked_prefill=True,
                 max_num_tokens=256,
-                use_cuda_graph=cuda_graph) as llm:
+                cuda_graph_config=CudaGraphConfig()
+                if cuda_graph else None) as llm:
             assert llm.args.quant_config.quant_algo == QuantAlgo.NVFP4
             assert llm.args.quant_config.kv_cache_quant_algo == QuantAlgo.FP8
             task = MMLU(self.MODEL_NAME)