[CI] refect e2e test (#4799)

### What this PR does / why we need it?
This PR updates the CI configuration and adjusts a set of end-to-end
(e2e) tests under tests/e2e/multicard, in order to refactor the test
suite and ensure compatibility with current codebase and CI workflows.

1. tests/e2e/multicard/test_prefix_caching.py: change model to Qwen3-8B
and rename the test case
2. tests/e2e/multicard/test_quantization.py: rename the test case
3. tests/e2e/multicard/test_qwen3_moe.py: remove duplicate test and
rename test cases
4. tests/e2e/multicard/test_qwen3_next.py: rename test cases and change
the W8A8 pruning model to the W8A8 model and remove the eager parameter
5. tests/e2e/multicard/test_shared_expert_dp.py: rename test case and
remove the eager parameter
6. tests/e2e/multicard/test_single_request_aclgraph.py: rename test case
and change Qwen3-30B to Qwen3-0.6B
7. tests/e2e/multicard/test_torchair_graph_mode.py: delete test cases
about torchair

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: hfadzxy <starmoon_zhang@163.com>
This commit is contained in:
zhangxinyuehfad
2025-12-12 08:42:08 +08:00
committed by GitHub
parent a6ef3ac4e4
commit bfafe30953
8 changed files with 30 additions and 66 deletions

View File

@@ -7,13 +7,13 @@ from tests.e2e.conftest import VllmRunner
from tests.e2e.model_utils import check_outputs_equal
MODELS = [
"vllm-ascend/DeepSeek-V2-Lite",
"deepseek-ai/DeepSeek-V2-Lite",
]
os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
@pytest.mark.parametrize("model", MODELS)
def test_models_with_enable_shared_expert_dp(model: str) -> None:
def test_deepseek_v2_lite_enable_shared_expert_dp_tp2(model: str) -> None:
if 'HCCL_OP_EXPANSION_MODE' in os.environ:
del os.environ['HCCL_OP_EXPANSION_MODE']
@@ -51,7 +51,7 @@ def test_models_with_enable_shared_expert_dp(model: str) -> None:
model,
max_model_len=1024,
tensor_parallel_size=2,
enforce_eager=False,
enable_expert_parallel=True,
compilation_config={
"cudagraph_capture_sizes": [1, 4, 8, 16],
"cudagraph_mode": "FULL_DECODE_ONLY",