[CI/UT][Refactor] move e2e spec decode and deepseek acc test to per pr

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

Author: MengqingCao

.github/workflows/vllm_ascend_test.yaml

@@ -265,7 +265,13 @@ jobs:
           --ignore=tests/e2e/singlecard/test_offline_inference.py \
           --ignore=tests/e2e/singlecard/test_ilama_lora.py \
           --ignore=tests/e2e/singlecard/test_guided_decoding.py \
-          --ignore=tests/e2e/singlecard/test_camem.py
+          --ignore=tests/e2e/singlecard/test_camem.py \
+          --ignore=tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py \
+          --ignore=tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
+          # ------------------------------------ v1 spec decode test ------------------------------------ #
+          VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py
+          # TODO: revert me when test_v1_spec_decode.py::test_ngram_correctness is fixed
+          VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
 
       - name: Run e2e test on V0 engine
         if: ${{ github.event_name == 'schedule' }}

.github/workflows/vllm_ascend_test_long_term.yaml

@@ -100,10 +100,6 @@ jobs:
             # v0 spec decode test
             VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/long_term/spec_decode_v0/e2e/test_mtp_correctness.py  # it needs a clean process
             pytest -sv tests/e2e/long_term/spec_decode_v0 --ignore=tests/e2e/long_term/spec_decode_v0/e2e/test_mtp_correctness.py
-            # v1 spec decode test
-            VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/long_term/spec_decode_v1/test_v1_mtp_correctness.py
-            # TODO: revert me when test_v1_spec_decode.py::test_ngram_correctness is fixed
-            VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/long_term/spec_decode_v1/test_v1_spec_decode.py
             # accuracy test single card
             pytest -sv tests/e2e/long_term/test_accuracy.py
           else

tests/e2e/long_term/test_deepseek_v2_lite_tp2_accuracy.py renamed to tests/e2e/multicard/test_deepseek_v2_lite_tp2_accuracy.py

@@ -23,11 +23,14 @@
 import os
 from unittest.mock import patch
 
+import pytest
+import vllm  # noqa: F401
 from modelscope import snapshot_download  # type: ignore
 from vllm import SamplingParams
 from vllm.model_executor.models.registry import ModelRegistry
 
 from tests.conftest import VllmRunner
+from tests.model_utils import check_outputs_equal
 
 os.environ["PYTORCH_NPU_ALLOC_CONF"] = "max_split_size_mb:256"
 
@@ -47,21 +50,6 @@ def test_models_distributed_QwQ():
         vllm_model.generate_greedy(example_prompts, max_tokens)
 
 
-def test_models_distributed_DeepSeek():
-    example_prompts = [
-        "Hello, my name is",
-    ]
-    dtype = "half"
-    max_tokens = 5
-    with VllmRunner(
-            "deepseek-ai/DeepSeek-V2-Lite",
-            dtype=dtype,
-            tensor_parallel_size=4,
-            distributed_executor_backend="mp",
-    ) as vllm_model:
-        vllm_model.generate_greedy(example_prompts, max_tokens)
-
-
 @patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_TOPK_OPTIMIZE": "1"})
 def test_models_distributed_topk() -> None:
     example_prompts = [
@@ -84,26 +72,6 @@ def test_models_distributed_topk() -> None:
         vllm_model.generate(example_prompts, sampling_params)
 
 
-@patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_DBO": "1"})
-def test_models_distributed_DeepSeek_dbo():
-    example_prompts = ["The president of the United States is"] * 41
-    dtype = "half"
-    sampling_params = SamplingParams(max_tokens=100, temperature=0.0)
-    with VllmRunner(
-            "deepseek-ai/DeepSeek-V2-Lite",
-            dtype=dtype,
-            tensor_parallel_size=4,
-            distributed_executor_backend="mp",
-    ) as vllm_model:
-        model_arch = 'DeepseekV2ForCausalLM'
-        registed_models = ModelRegistry.models
-        assert registed_models[
-            model_arch].module_name == "vllm_ascend.models.deepseek_dbo"
-        assert registed_models[
-            model_arch].class_name == "CustomDeepseekDBOForCausalLM"
-        vllm_model.generate(example_prompts, sampling_params)
-
-
 @patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_DBO": "1"})
 def test_models_distributed_DeepSeekV3_dbo():
     example_prompts = ["The president of the United States is"] * 41
@@ -139,3 +107,36 @@ def test_models_distributed_DeepSeek_W8A8():
             quantization="ascend",
     ) as vllm_model:
         vllm_model.generate_greedy(example_prompts, max_tokens)
+
+
+def test_models_distributed_DeepSeek_dbo(monkeypatch: pytest.MonkeyPatch):
+    with monkeypatch.context() as m:
+        m.setenv("VLLM_ASCEND_ENABLE_DBO", "1")
+
+        example_prompts = ["The president of the United States is"] * 41
+        dtype = "half"
+        sampling_params = SamplingParams(max_tokens=100, temperature=0.0)
+        with VllmRunner(
+                "deepseek-ai/DeepSeek-V2-Lite",
+                dtype=dtype,
+                tensor_parallel_size=4,
+                distributed_executor_backend="mp",
+        ) as vllm_model:
+            dpo_output = vllm_model.generate(example_prompts, sampling_params)
+
+    with monkeypatch.context() as m:
+        m.setenv("VLLM_ASCEND_ENABLE_DBO", "0")
+        with VllmRunner(
+                "deepseek-ai/DeepSeek-V2-Lite",
+                dtype=dtype,
+                tensor_parallel_size=4,
+                distributed_executor_backend="mp",
+        ) as vllm_model:
+            output = vllm_model.generate(example_prompts, sampling_params)
+
+    check_outputs_equal(
+        outputs_0_lst=output,
+        outputs_1_lst=dpo_output,
+        name_0="vllm_outputs",
+        name_1="vllm_dbo_outputs",
+    )