Refactor e2e CI (#2276)

Refactor E2E CI to make it clear and faster
1. remove some uesless e2e test
2. remove some uesless function
3. Make sure all test runs with VLLMRunner to avoid oom error
4. Make sure all ops test end with torch.empty_cache to avoid oom error
5. run the test one by one to avoid resource limit error


- vLLM version: v0.10.1.1
- vLLM main:
a344a5aa0a

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

tests/e2e/singlecard/test_ascend_scheduler.py

@@ -4,6 +4,7 @@ import pytest
 
 from tests.e2e.conftest import VllmRunner
 from tests.e2e.model_utils import check_outputs_equal
+from vllm_ascend.ascend_config import clear_ascend_config
 
 MODEL = "Qwen/Qwen3-0.6B"
 
@@ -26,6 +27,8 @@ def test_concurrent_partial_prefill():
         for output in outputs:
             assert len(output.outputs) == 1
 
+    clear_ascend_config()
+
 
 def test_prefix_cache_stats_is_recorded():
     with VllmRunner(MODEL,
@@ -45,13 +48,17 @@ def test_prefix_cache_stats_is_recorded():
         outputs = vllm_model.model.generate([input_tokens])
         assert outputs[0].num_cached_tokens == 128
 
+    clear_ascend_config()
+
 
 @pytest.mark.parametrize("max_tokens",
                          [4])  # cannot align results when max_tokens > 4
 @pytest.mark.parametrize("chunked_prefill_token_size", [16])
 def test_chunked_prefill_with_ascend_scheduler(
-        example_prompts, max_tokens: int,
-        chunked_prefill_token_size: int) -> None:
+        max_tokens: int, chunked_prefill_token_size: int) -> None:
+    example_prompts = [
+        "vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs."
+    ]
     max_num_seqs = chunked_prefill_token_size
     max_num_batched_tokens = chunked_prefill_token_size
     with VllmRunner(MODEL,
@@ -63,7 +70,6 @@ def test_chunked_prefill_with_ascend_scheduler(
                     },
                     max_num_seqs=max_num_seqs,
                     max_num_batched_tokens=max_num_batched_tokens,
-                    enforce_eager=True,
                     max_model_len=2048,
                     gpu_memory_utilization=0.7) as vllm_model:
         chunked_prefill_output = vllm_model.generate_greedy(
@@ -75,7 +81,6 @@ def test_chunked_prefill_with_ascend_scheduler(
                             'enabled': True,
                         },
                     },
-                    enforce_eager=True,
                     max_model_len=2048,
                     gpu_memory_utilization=0.7) as vllm_model:
         vllm_output = vllm_model.generate_greedy(example_prompts, max_tokens)
@@ -86,3 +91,4 @@ def test_chunked_prefill_with_ascend_scheduler(
         name_0="vllm_output",
         name_1="chunked_prefill_output",
     )
+    clear_ascend_config()