[CI] fix vllm test (#365)

fix vllm test

Signed-off-by: MengqingCao <cmq0113@163.com>

.github/workflows/vllm_ascend_test.yaml

@@ -130,11 +130,12 @@ jobs:
         run: |
           VLLM_USE_V1=0 pytest -sv tests
 
-      # FIXME: make vllm test pass
-      #- name: Checkout vllm-project/vllm repo
-      #- name: Run vllm-project/vllm test
-      #  run: |
-      #    VLLM_USE_V1=0 pytest -sv
+      - name: Run vllm-project/vllm test
+        env:
+          VLLM_USE_V1: 0
+          PYTORCH_NPU_ALLOC_CONF: max_split_size_mb:256
+        run: |
+          pytest -sv
 
   post_cleanup:
     name: vLLM Ascend test (post-cleanup)

pytest.ini

@@ -53,6 +53,11 @@ addopts = --ignore=vllm-empty/tests/test_utils.py
           --ignore=vllm-empty/tests/multimodal/test_processing.py
           --ignore=vllm-empty/tests/multimodal/test_processor_kwargs.py
           --ignore=vllm-empty/tests/multimodal/test_utils.py
+        ;   Both V1 and V0 engine will be run in detokenizer/test_stop_reason
+        ;   VLLM_USE_V1=1 is not supported with device type=npu.
+          --ignore=vllm-empty/tests/detokenizer/test_stop_reason.py
+        ;   oom on llama-2-7b-hf
+          --ignore=vllm-empty/tests/detokenizer/test_stop_strings.py
 
 testpaths =
     vllm-empty/tests