diff --git a/.github/workflows/_e2e_test.yaml b/.github/workflows/_e2e_test.yaml
index f3edf491..5ee2fd25 100644
--- a/.github/workflows/_e2e_test.yaml
+++ b/.github/workflows/_e2e_test.yaml
@@ -92,7 +92,7 @@ jobs:
           # pytest -sv --durations=0 tests/e2e/singlecard/test_aclgraph_accuracy.py
           # pytest -sv --durations=0 tests/e2e/singlecard/test_quantization.py
           pytest -sv --durations=0 tests/e2e/singlecard/test_vlm.py::test_multimodal_vl
-          pytest -sv --durations=0 tests/e2e/singlecard/pooling/test_classification.py::test_classify_correctness
+          pytest -sv --durations=0 tests/e2e/singlecard/pooling/test_classification.py::test_qwen_pooling_classify_correctness
 
       - name: Run e2e test
         env:
diff --git a/tests/e2e/multicard/long_sequence/test_accuracy.py b/tests/e2e/multicard/long_sequence/test_accuracy.py
index a8bbf50a..af111899 100644
--- a/tests/e2e/multicard/long_sequence/test_accuracy.py
+++ b/tests/e2e/multicard/long_sequence/test_accuracy.py
@@ -36,7 +36,7 @@ MODELS = [
                     reason="0.12.0 is not supported for context sequence.")
 @pytest.mark.parametrize("model", MODELS)
 @pytest.mark.parametrize("max_tokens", [10])
-def test_output_between_tp_and_cp(
+def test_models_long_sequence_output_between_tp_and_cp(
     model: str,
     max_tokens: int,
 ) -> None:
@@ -69,7 +69,6 @@ def test_output_between_tp_and_cp(
             "tensor_parallel_size": 1,
             "decode_context_parallel_size": 1,
             "prefill_context_parallel_size": 2,
-            "enforce_eager": False,
             "compilation_config": {
                 "cudagraph_mode": "FULL_DECODE_ONLY",
                 "cudagraph_capture_sizes": [4, 8, 24, 48, 60]
diff --git a/tests/e2e/multicard/long_sequence/test_basic.py b/tests/e2e/multicard/long_sequence/test_basic.py
index 25422678..bc6f839a 100644
--- a/tests/e2e/multicard/long_sequence/test_basic.py
+++ b/tests/e2e/multicard/long_sequence/test_basic.py
@@ -34,7 +34,7 @@ os.environ["HCCL_BUFFSIZE"] = "768"
 
 @pytest.mark.skipif(vllm_version_is('0.12.0'),
                     reason="0.12.0 is not supported for context sequence.")
-def test_pcp_dcp_basic():
+def test_models_pcp_dcp_basic():
     prompts = [
         "The capital of France is", "Hello, my name is Tom, I am",
         "The president of United States is", "AI future is"
@@ -69,7 +69,7 @@ def test_pcp_dcp_basic():
 
 @pytest.mark.skipif(vllm_version_is('0.12.0'),
                     reason="0.12.0 is not supported for context sequence.")
-def test_pcp_dcp_full_graph():
+def test_models_pcp_dcp_full_graph():
     prompts = [
         "The capital of France is", "Hello, my name is Tom, I am",
         "The president of United States is", "AI future is"
@@ -77,7 +77,6 @@ def test_pcp_dcp_full_graph():
     model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
     sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
     with VllmRunner(model,
-                    enforce_eager=False,
                     max_model_len=1024,
                     tensor_parallel_size=2,
                     prefill_context_parallel_size=2,
@@ -93,7 +92,6 @@ def test_pcp_dcp_full_graph():
 
     model = "vllm-ascend/Qwen3-30B-A3B-W8A8"
     with VllmRunner(model,
-                    enforce_eager=False,
                     max_model_len=1024,
                     tensor_parallel_size=2,
                     prefill_context_parallel_size=2,
@@ -110,7 +108,7 @@ def test_pcp_dcp_full_graph():
 
 @pytest.mark.skipif(vllm_version_is('0.12.0'),
                     reason="0.12.0 is not supported for context sequence.")
-def test_pcp_dcp_piece_wise():
+def test_models_pcp_dcp_piece_wise():
     prompts = [
         "The capital of France is", "Hello, my name is Tom, I am",
         "The president of United States is", "AI future is"
@@ -118,7 +116,6 @@ def test_pcp_dcp_piece_wise():
     model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
     sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
     with VllmRunner(model,
-                    enforce_eager=False,
                     max_model_len=1024,
                     tensor_parallel_size=2,
                     prefill_context_parallel_size=2,
@@ -130,7 +127,6 @@ def test_pcp_dcp_piece_wise():
 
     model = "vllm-ascend/Qwen3-30B-A3B-W8A8"
     with VllmRunner(model,
-                    enforce_eager=False,
                     max_model_len=1024,
                     tensor_parallel_size=2,
                     prefill_context_parallel_size=2,
diff --git a/tests/e2e/multicard/test_expert_parallel.py b/tests/e2e/multicard/test_expert_parallel.py
index 762ca6d2..9149d12e 100644
--- a/tests/e2e/multicard/test_expert_parallel.py
+++ b/tests/e2e/multicard/test_expert_parallel.py
@@ -15,14 +15,12 @@ def test_deepseek_correctness_ep(model_name):
     max_tokens = 5
 
     # FIXME: Really strange that chunked prefill might lead to different results, investigate further
-    with VllmRunner(model_name, tensor_parallel_size=2,
-                    enforce_eager=False) as vllm_model:
+    with VllmRunner(model_name, tensor_parallel_size=2) as vllm_model:
         tp_output = vllm_model.generate_greedy(example_prompts, max_tokens)
 
     with VllmRunner(model_name,
                     tensor_parallel_size=2,
-                    enable_expert_parallel=True,
-                    enforce_eager=False) as vllm_model:
+                    enable_expert_parallel=True) as vllm_model:
         ep_output = vllm_model.generate_greedy(example_prompts, max_tokens)
 
     check_outputs_equal(
diff --git a/tests/e2e/multicard/test_full_graph_mode.py b/tests/e2e/multicard/test_full_graph_mode.py
index 362f8be7..e47b4e18 100644
--- a/tests/e2e/multicard/test_full_graph_mode.py
+++ b/tests/e2e/multicard/test_full_graph_mode.py
@@ -41,7 +41,6 @@ def test_qwen3_moe_full_decode_only_tp2():
     with VllmRunner(model,
                     max_model_len=1024,
                     tensor_parallel_size=2,
-                    enforce_eager=False,
                     compilation_config={
                         "cudagraph_mode": "FULL_DECODE_ONLY",
                         "cudagraph_capture_sizes": [4, 8, 24, 48, 60]
@@ -53,7 +52,6 @@ def test_qwen3_moe_full_decode_only_tp2():
             model,
             max_model_len=1024,
             tensor_parallel_size=2,
-            enforce_eager=False,
     ) as runner:
         vllm_eager_outputs = runner.model.generate(prompts, sampling_params)
 
@@ -87,7 +85,6 @@ def test_qwen3_moe_full_graph_tp2():
     with VllmRunner(model,
                     max_model_len=1024,
                     tensor_parallel_size=2,
-                    enforce_eager=False,
                     compilation_config={
                         "cudagraph_mode": "FULL",
                         "cudagraph_capture_sizes": [4, 8, 24, 48, 60]
@@ -99,7 +96,6 @@ def test_qwen3_moe_full_graph_tp2():
             model,
             max_model_len=1024,
             tensor_parallel_size=2,
-            enforce_eager=False,
     ) as runner:
         vllm_eager_outputs = runner.model.generate(prompts, sampling_params)
 
diff --git a/tests/e2e/multicard/test_ilama_lora_tp2.py b/tests/e2e/multicard/test_ilama_lora_tp2.py
index 750039c3..230aac16 100644
--- a/tests/e2e/multicard/test_ilama_lora_tp2.py
+++ b/tests/e2e/multicard/test_ilama_lora_tp2.py
@@ -8,15 +8,16 @@ from tests.e2e.singlecard.test_ilama_lora import (EXPECTED_LORA_OUTPUT,
 
 @pytest.mark.parametrize("distributed_executor_backend", ["mp"])
 def test_ilama_lora_tp2(distributed_executor_backend, ilama_lora_files):
-    with VllmRunner(snapshot_download(MODEL_PATH),
-                    enable_lora=True,
-                    max_loras=4,
-                    dtype="half",
-                    max_model_len=1024,
-                    max_num_seqs=16,
-                    tensor_parallel_size=2,
-                    distributed_executor_backend=distributed_executor_backend,
-                    enforce_eager=False) as vllm_model:
+    with VllmRunner(
+            snapshot_download(MODEL_PATH),
+            enable_lora=True,
+            max_loras=4,
+            dtype="half",
+            max_model_len=1024,
+            max_num_seqs=16,
+            tensor_parallel_size=2,
+            distributed_executor_backend=distributed_executor_backend,
+    ) as vllm_model:
         output = do_sample(vllm_model.model, ilama_lora_files, lora_id=2)
 
     for i in range(len(EXPECTED_LORA_OUTPUT)):
diff --git a/tests/e2e/multicard/test_offline_inference_distributed.py b/tests/e2e/multicard/test_offline_inference_distributed.py
index 558067ce..e5f4b2c2 100644
--- a/tests/e2e/multicard/test_offline_inference_distributed.py
+++ b/tests/e2e/multicard/test_offline_inference_distributed.py
@@ -189,7 +189,6 @@ def test_qwen3_dense_fc1_tp2(model):
     with VllmRunner(
             snapshot_download(model),
             max_model_len=8192,
-            enforce_eager=False,
             dtype="auto",
             tensor_parallel_size=2,
             quantization="ascend",
@@ -209,7 +208,6 @@ def test_qwen3_dense_prefetch_mlp_weight_tp2(model):
     with VllmRunner(
             snapshot_download(model),
             max_model_len=8192,
-            enforce_eager=False,
             dtype="auto",
             tensor_parallel_size=2,
             quantization="ascend",
diff --git a/tests/e2e/singlecard/pooling/test_classification.py b/tests/e2e/singlecard/pooling/test_classification.py
index e59983c1..8bdd3660 100644
--- a/tests/e2e/singlecard/pooling/test_classification.py
+++ b/tests/e2e/singlecard/pooling/test_classification.py
@@ -5,7 +5,7 @@ from transformers import AutoModelForSequenceClassification
 from tests.e2e.conftest import HfRunner, VllmRunner
 
 
-def test_classify_correctness() -> None:
+def test_qwen_pooling_classify_correctness() -> None:
 
     model_name = snapshot_download("Howeee/Qwen2.5-1.5B-apeach")
 
diff --git a/tests/e2e/singlecard/pooling/test_embedding.py b/tests/e2e/singlecard/pooling/test_embedding.py
index a564dfbb..8800fc7a 100644
--- a/tests/e2e/singlecard/pooling/test_embedding.py
+++ b/tests/e2e/singlecard/pooling/test_embedding.py
@@ -36,7 +36,6 @@ def test_embed_models_correctness(model: str):
     with VllmRunner(
             model_name,
             runner="pooling",
-            enforce_eager=False,
             max_model_len=None,
             cudagraph_capture_sizes=[4],
     ) as vllm_runner:
@@ -58,14 +57,13 @@ def test_embed_models_correctness(model: str):
     )
 
 
-def test_bge_model_correctness():
+def test_bge_m3_correctness():
     queries = ['What is the capital of China?', 'Explain gravity']
 
     model_name = snapshot_download("BAAI/bge-m3")
     with VllmRunner(
             model_name,
             runner="pooling",
-            enforce_eager=False,
     ) as vllm_aclgraph_runner:
         vllm_aclgraph_outputs = vllm_aclgraph_runner.embed(queries)
 
diff --git a/tests/e2e/singlecard/pooling/test_scoring.py b/tests/e2e/singlecard/pooling/test_scoring.py
index c196a0bf..9656b422 100644
--- a/tests/e2e/singlecard/pooling/test_scoring.py
+++ b/tests/e2e/singlecard/pooling/test_scoring.py
@@ -34,7 +34,7 @@ def model_name(request):
     yield snapshot_download(request.param)
 
 
-def test_cross_encoder_1_to_1(model_name):
+def test_cross_encoder_score_1_to_1(model_name):
     text_pair = [TEXTS_1[0], TEXTS_2[0]]
 
     with HfRunner(model_name, dtype=DTYPE, is_cross_encoder=True) as hf_model:
@@ -53,7 +53,7 @@ def test_cross_encoder_1_to_1(model_name):
     assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
 
 
-def test_cross_encoder_1_to_N(model_name):
+def test_cross_encoder_score_1_to_N(model_name):
     text_pairs = [
         [TEXTS_1[0], TEXTS_2[0]],
         [TEXTS_1[0], TEXTS_2[1]],
@@ -76,7 +76,7 @@ def test_cross_encoder_1_to_N(model_name):
     assert hf_outputs[1] == pytest.approx(vllm_outputs[1], rel=0.01)
 
 
-def test_cross_encoder_N_to_N(model_name):
+def test_cross_encoder_score_N_to_N(model_name):
     text_pairs = [
         [TEXTS_1[0], TEXTS_2[0]],
         [TEXTS_1[1], TEXTS_2[1]],
@@ -104,7 +104,7 @@ def emb_model_name(request):
     yield snapshot_download(request.param)
 
 
-def test_embedding_1_to_1(emb_model_name):
+def test_embedding_score_1_to_1(emb_model_name):
     text_pair = [TEXTS_1[0], TEXTS_2[0]]
 
     with HfRunner(emb_model_name, dtype=DTYPE,
@@ -127,7 +127,7 @@ def test_embedding_1_to_1(emb_model_name):
     assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
 
 
-def test_embedding_1_to_N(emb_model_name):
+def test_embedding_score_1_to_N(emb_model_name):
     text_pairs = [
         [TEXTS_1[0], TEXTS_2[0]],
         [TEXTS_1[0], TEXTS_2[1]],
@@ -157,7 +157,7 @@ def test_embedding_1_to_N(emb_model_name):
     assert hf_outputs[1] == pytest.approx(vllm_outputs[1], rel=0.01)
 
 
-def test_embedding_N_to_N(emb_model_name):
+def test_embedding_score_N_to_N(emb_model_name):
     text_pairs = [
         [TEXTS_1[0], TEXTS_2[0]],
         [TEXTS_1[1], TEXTS_2[1]],
diff --git a/tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py b/tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py
index 20f7d145..d8c8fb23 100644
--- a/tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py
+++ b/tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py
@@ -171,4 +171,4 @@ def test_mtp2_correctness_piecewise_graph_with_pad(
     mtp_correctness(sampling_config,
                     model_name,
                     2,
-                    disable_padded_drafter_batch=False)
+                    disable_padded_drafter_batch=False)
\ No newline at end of file
diff --git a/tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py b/tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
index e42e4bd1..1526997a 100644
--- a/tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
+++ b/tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
@@ -76,19 +76,22 @@ def test_ngram_correctness(
     should be the same when using ngram speculative decoding.
     '''
 
-    with VllmRunner(model_name, max_model_len=1024,
-                    enforce_eager=False) as ref_llm:
+    with VllmRunner(
+            model_name,
+            max_model_len=1024,
+    ) as ref_llm:
         ref_outputs = ref_llm.model.chat(test_prompts, sampling_config)
 
-    with VllmRunner(model_name,
-                    speculative_config={
-                        "method": "ngram",
-                        "prompt_lookup_max": 5,
-                        "prompt_lookup_min": 3,
-                        "num_speculative_tokens": 3,
-                    },
-                    max_model_len=1024,
-                    enforce_eager=False) as runner:
+    with VllmRunner(
+            model_name,
+            speculative_config={
+                "method": "ngram",
+                "prompt_lookup_max": 5,
+                "prompt_lookup_min": 3,
+                "num_speculative_tokens": 3,
+            },
+            max_model_len=1024,
+    ) as runner:
         spec_outputs = runner.model.chat(test_prompts, sampling_config)
     matches = 0
     misses = 0
@@ -190,8 +193,7 @@ def test_suffix_correctness(
     Compare the outputs of a original LLM and a speculative LLM
     should be the same when using ngram speculative decoding.
     '''
-    with VllmRunner(model_name, max_model_len=1024,
-                    enforce_eager=False) as ref_llm:
+    with VllmRunner(model_name, max_model_len=1024) as ref_llm:
         ref_outputs = ref_llm.model.chat(test_prompts, sampling_config)
 
     with VllmRunner(model_name,
@@ -199,8 +201,7 @@ def test_suffix_correctness(
                         "method": "suffix",
                         "num_speculative_tokens": 8,
                     },
-                    max_model_len=1024,
-                    enforce_eager=False) as runner:
+                    max_model_len=1024) as runner:
         spec_outputs = runner.model.chat(test_prompts, sampling_config)
     matches = 0
     misses = 0
@@ -236,8 +237,7 @@ def test_suffix_acceptance(
                         "num_speculative_tokens": 10,
                     },
                     max_model_len=1024,
-                    disable_log_stats=False,
-                    enforce_eager=False) as runner:
+                    disable_log_stats=False) as runner:
         for i in range(10):
             runner.model.chat(test_prompts[i], sampling_config)
             metrics = runner.model.get_metrics()
@@ -278,7 +278,7 @@ def test_eagle_logprobs(
                                      max_tokens=10,
                                      ignore_eos=False)
 
-    ref_llm = LLM(model=model_name, max_model_len=2048, enforce_eager=False)
+    ref_llm = LLM(model=model_name, max_model_len=2048)
     ref_outputs = ref_llm.chat([prompt], sampling_params)
     ref_logprobs = []
     for output in ref_outputs[0].outputs:
@@ -300,7 +300,6 @@ def test_eagle_logprobs(
                 "max_model_len": 128,
             },
             max_model_len=128,
-            enforce_eager=False,
     ) as runner:
         spec_outputs = runner.model.chat([prompt], sampling_params)
 
diff --git a/tests/e2e/singlecard/test_aclgraph_accuracy.py b/tests/e2e/singlecard/test_aclgraph_accuracy.py
index 9eb68894..baba92ea 100644
--- a/tests/e2e/singlecard/test_aclgraph_accuracy.py
+++ b/tests/e2e/singlecard/test_aclgraph_accuracy.py
@@ -36,7 +36,7 @@ MODELS = [
 
 @pytest.mark.parametrize("model", MODELS)
 @pytest.mark.parametrize("max_tokens", [32])
-def test_output_between_eager_and_aclgraph(
+def test_models_output_between_eager_and_aclgraph(
     model: str,
     max_tokens: int,
 ) -> None:
@@ -50,7 +50,6 @@ def test_output_between_eager_and_aclgraph(
         with VllmRunner(
                 model,
                 max_model_len=1024,
-                enforce_eager=False,
                 quantization="ascend",
         ) as runner:
             vllm_aclgraph_outputs = runner.model.generate(
@@ -68,7 +67,6 @@ def test_output_between_eager_and_aclgraph(
         with VllmRunner(
                 model,
                 max_model_len=1024,
-                enforce_eager=False,
         ) as runner:
             vllm_aclgraph_outputs = runner.model.generate(
                 prompts, sampling_params)
@@ -100,7 +98,7 @@ def test_output_between_eager_and_aclgraph(
 
 @pytest.mark.parametrize("model", MODELS)
 @pytest.mark.parametrize("max_tokens", [32])
-def test_output_between_eager_and_full_decode_only(
+def test_models_output_between_eager_and_full_decode_only(
     model: str,
     max_tokens: int,
 ) -> None:
@@ -155,7 +153,6 @@ def test_output_between_eager_and_full_decode_only(
         with VllmRunner(
                 model,
                 max_model_len=1024,
-                enforce_eager=False,
                 compilation_config={"cudagraph_mode": "FULL_DECODE_ONLY"},
                 quantization="ascend",
         ) as runner:
@@ -166,7 +163,6 @@ def test_output_between_eager_and_full_decode_only(
         with VllmRunner(
                 model,
                 max_model_len=1024,
-                enforce_eager=False,
                 compilation_config={
                     "cudagraph_capture_sizes": [4, 8, 32, 64],
                     "cudagraph_mode": "FULL_DECODE_ONLY"
@@ -196,7 +192,7 @@ def test_output_between_eager_and_full_decode_only(
 
 @pytest.mark.parametrize("model", MODELS)
 @pytest.mark.parametrize("max_tokens", [32])
-def test_output_between_eager_and_fullgraph_npugraph_ex(
+def test_models_output_between_eager_and_fullgraph_npugraph_ex(
     model: str,
     max_tokens: int,
 ) -> None:
@@ -251,7 +247,6 @@ def test_output_between_eager_and_fullgraph_npugraph_ex(
         with VllmRunner(
                 model,
                 max_model_len=1024,
-                enforce_eager=False,
                 compilation_config={"cudagraph_mode": "FULL_DECODE_ONLY"},
                 additional_config={"enable_npugraph_ex": True},
                 quantization="ascend",
@@ -263,7 +258,6 @@ def test_output_between_eager_and_fullgraph_npugraph_ex(
         with VllmRunner(
                 model,
                 max_model_len=1024,
-                enforce_eager=False,
                 compilation_config={
                     "cudagraph_capture_sizes": [4, 8, 32, 64],
                     "cudagraph_mode": "FULL_DECODE_ONLY"
diff --git a/tests/e2e/singlecard/test_camem.py b/tests/e2e/singlecard/test_camem.py
index cdf7527e..33a99a42 100644
--- a/tests/e2e/singlecard/test_camem.py
+++ b/tests/e2e/singlecard/test_camem.py
@@ -76,9 +76,7 @@ def test_end_to_end():
     prompt = "How are you?"
     sampling_params = SamplingParams(temperature=0, max_tokens=10)
 
-    with VllmRunner("Qwen/Qwen3-0.6B",
-                    enforce_eager=False,
-                    enable_sleep_mode=True) as runner:
+    with VllmRunner("Qwen/Qwen3-0.6B", enable_sleep_mode=True) as runner:
 
         output = runner.model.generate(prompt, sampling_params)
         # the benefit of `llm.sleep(level=2)` is mainly CPU memory usage,
diff --git a/tests/e2e/singlecard/test_ilama_lora.py b/tests/e2e/singlecard/test_ilama_lora.py
index daeac4aa..1b941a06 100644
--- a/tests/e2e/singlecard/test_ilama_lora.py
+++ b/tests/e2e/singlecard/test_ilama_lora.py
@@ -45,13 +45,14 @@ def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> list[str]:
 
 
 def test_ilama_lora(ilama_lora_files):
-    with VllmRunner(snapshot_download(MODEL_PATH),
-                    enable_lora=True,
-                    dtype="half",
-                    max_loras=4,
-                    max_model_len=1024,
-                    max_num_seqs=16,
-                    enforce_eager=False) as vllm_model:
+    with VllmRunner(
+            snapshot_download(MODEL_PATH),
+            enable_lora=True,
+            dtype="half",
+            max_loras=4,
+            max_model_len=1024,
+            max_num_seqs=16,
+    ) as vllm_model:
 
         output1 = do_sample(vllm_model.model, ilama_lora_files, lora_id=1)
         for i in range(len(EXPECTED_LORA_OUTPUT)):
diff --git a/tests/e2e/singlecard/test_multistream_overlap_shared_expert.py b/tests/e2e/singlecard/test_multistream_overlap_shared_expert.py
index 89a78912..eb00cc67 100644
--- a/tests/e2e/singlecard/test_multistream_overlap_shared_expert.py
+++ b/tests/e2e/singlecard/test_multistream_overlap_shared_expert.py
@@ -58,7 +58,6 @@ def test_models_with_multistream_overlap_shared_expert(
     with VllmRunner(
             model,
             max_model_len=1024,
-            enforce_eager=False,
             additional_config={
                 "multistream_overlap_shared_expert": True,
             },
diff --git a/tests/e2e/singlecard/test_quantization.py b/tests/e2e/singlecard/test_quantization.py
index 95f26ee8..6ab54084 100644
--- a/tests/e2e/singlecard/test_quantization.py
+++ b/tests/e2e/singlecard/test_quantization.py
@@ -20,15 +20,14 @@ from modelscope import snapshot_download  # type: ignore[import-untyped]
 from tests.e2e.conftest import VllmRunner
 
 
-def test_quant_W8A8():
+def test_qwen3_w8a8_quant():
     max_tokens = 5
     example_prompts = [
         "vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs."
     ]
     with VllmRunner(
-            snapshot_download("vllm-ascend/Qwen2.5-0.5B-Instruct-W8A8"),
+            snapshot_download("vllm-ascend/Qwen3-0.6B-W8A8"),
             max_model_len=8192,
-            enforce_eager=False,
             gpu_memory_utilization=0.7,
             quantization="ascend",
     ) as vllm_model:
diff --git a/tests/e2e/singlecard/test_sampler.py b/tests/e2e/singlecard/test_sampler.py
index 73055f3a..fbb03913 100644
--- a/tests/e2e/singlecard/test_sampler.py
+++ b/tests/e2e/singlecard/test_sampler.py
@@ -21,7 +21,7 @@ from vllm import SamplingParams
 from tests.e2e.conftest import VllmRunner
 
 
-def test_models_topk() -> None:
+def test_qwen3_topk() -> None:
     example_prompts = [
         "Hello, my name is",
     ]
@@ -36,7 +36,7 @@ def test_models_topk() -> None:
         runner.generate(example_prompts, sampling_params)
 
 
-def test_models_prompt_logprobs() -> None:
+def test_qwen3_prompt_logprobs() -> None:
     example_prompts = [
         "Hello, my name is",
     ]
@@ -49,7 +49,7 @@ def test_models_prompt_logprobs() -> None:
                                         num_logprobs=1)
 
 
-def test_exponential_overlap() -> None:
+def test_qwen3_exponential_overlap() -> None:
     example_prompts = [
         "Hello, my name is",
     ]
diff --git a/tests/e2e/singlecard/test_vlm.py b/tests/e2e/singlecard/test_vlm.py
index 6133157f..d2ca42be 100644
--- a/tests/e2e/singlecard/test_vlm.py
+++ b/tests/e2e/singlecard/test_vlm.py
@@ -46,7 +46,6 @@ def test_multimodal_vl(vl_config):
 
     with VllmRunner(vl_config["model"],
                     mm_processor_kwargs=vl_config["mm_processor_kwargs"],
-                    enforce_eager=False,
                     max_model_len=8192,
                     limit_mm_per_prompt={"image": 1}) as vllm_model:
         outputs = vllm_model.generate_greedy(
diff --git a/tests/e2e/singlecard/test_xlite.py b/tests/e2e/singlecard/test_xlite.py
index 0b10e63e..eba65a99 100644
--- a/tests/e2e/singlecard/test_xlite.py
+++ b/tests/e2e/singlecard/test_xlite.py
@@ -48,7 +48,6 @@ def test_models_with_xlite_decode_only(
             model,
             block_size=128,
             max_model_len=1024,
-            enforce_eager=False,
             additional_config={"xlite_graph_config": {
                 "enabled": True
             }},
@@ -97,7 +96,6 @@ def test_models_with_xlite_full_mode(
             model,
             block_size=128,
             max_model_len=1024,
-            enforce_eager=False,
             additional_config={
                 "xlite_graph_config": {
                     "enabled": True,