diff --git a/examples/pytorch/quickstart_advanced.py b/examples/pytorch/quickstart_advanced.py
index 59a6cbd86c..f5b25da0a5 100644
--- a/examples/pytorch/quickstart_advanced.py
+++ b/examples/pytorch/quickstart_advanced.py
@@ -78,6 +78,13 @@ def add_llm_args(parser):
                         default=False,
                         action='store_true')
     parser.add_argument('--use_cuda_graph', default=False, action='store_true')
+    parser.add_argument('--cuda_graph_padding_enabled',
+                        default=False,
+                        action='store_true')
+    parser.add_argument('--cuda_graph_batch_sizes',
+                        nargs='+',
+                        type=int,
+                        default=[])
     parser.add_argument('--print_iter_log',
                         default=False,
                         action='store_true',
@@ -160,6 +167,8 @@ def setup_llm(args):
               kv_cache_config=kv_cache_config,
               attn_backend=args.attention_backend,
               use_cuda_graph=args.use_cuda_graph,
+              cuda_graph_padding_enabled=args.cuda_graph_padding_enabled,
+              cuda_graph_batch_sizes=args.cuda_graph_batch_sizes,
               load_format=args.load_format,
               print_iter_log=args.print_iter_log,
               enable_iter_perf_stats=args.print_iter_log,
diff --git a/tensorrt_llm/_torch/pyexecutor/model_engine.py b/tensorrt_llm/_torch/pyexecutor/model_engine.py
index 49bf6194b2..c6cbc47bfb 100644
--- a/tensorrt_llm/_torch/pyexecutor/model_engine.py
+++ b/tensorrt_llm/_torch/pyexecutor/model_engine.py
@@ -478,6 +478,10 @@ class PyTorchModelEngine(ModelEngine):
             logger.info("Skipping warm up as no KV Cache manager allocated.")
             return
 
+        # The lifetime of model engine and kv cache manager can be different.
+        # Reset the global cuda graph dummy request to None in warmup.
+        self.cuda_graph_dummy_request = None
+
         def get_cuda_graph_warmup_request(batch_size):
             available_blocks = kv_cache_manager.get_num_free_blocks()
             if available_blocks >= batch_size:
diff --git a/tests/integration/defs/test_e2e.py b/tests/integration/defs/test_e2e.py
index c907ae0ac7..943194f94a 100644
--- a/tests/integration/defs/test_e2e.py
+++ b/tests/integration/defs/test_e2e.py
@@ -1569,6 +1569,31 @@ def test_ptq_quickstart_advanced_mtp(llm_root, llm_venv, model_name,
         _check_mem_usage(running_log, [54.50, 0, 0, 0])
 
 
+@pytest.mark.skip_less_device(4)
+def test_ptq_quickstart_advanced_bs1(llm_root, llm_venv):
+    model_name = "DeepSeek-V3-Lite-FP8"
+    model_path = "DeepSeek-V3-Lite/fp8"
+    print(f"Testing {model_name}.")
+    example_root = Path(os.path.join(llm_root, "examples", "pytorch"))
+    llm_venv.run_cmd([
+        str(example_root / "quickstart_advanced.py"),
+        "--use_cuda_graph",
+        "--cuda_graph_padding_enabled",
+        "--cuda_graph_batch_sizes",
+        "8",
+        "--disable_overlap_scheduler",
+        "--enable_attention_dp",
+        "--tp_size",
+        "4",
+        "--moe_ep_size",
+        "4",
+        "--prompt",
+        "\"NVIDIA is a great company because\"",
+        "--model_dir",
+        f"{llm_models_root()}/{model_path}",
+    ])
+
+
 @pytest.mark.skip_less_device_memory(80000)
 @pytest.mark.skip_less_device(8)
 @skip_pre_hopper
diff --git a/tests/integration/test_lists/test-db/l0_dgx_h100.yml b/tests/integration/test_lists/test-db/l0_dgx_h100.yml
index e1b96b68c1..b319a30df3 100644
--- a/tests/integration/test_lists/test-db/l0_dgx_h100.yml
+++ b/tests/integration/test_lists/test-db/l0_dgx_h100.yml
@@ -53,6 +53,7 @@ l0_dgx_h100:
       auto_trigger: deepseek
   tests:
   - unittest/_torch/multi_gpu_modeling -k "deepseek"
+  - test_e2e.py::test_ptq_quickstart_advanced_bs1
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]