[Misc] Clean up useless patch (#3320)

### What this PR does / why we need it?
1. clean up v0.10.2 support in ut and e2e test
2. remove v0.11.0 period job, we're at v0.11.0 now.
3. remove uesless patch for deepseek v3.2. They have been done in vLLM
already.
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?


- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

Signed-off-by: wangxiyuan <[email protected]>

Author: wangxiyuan

.github/workflows/vllm_ascend_test_full_vllm_0.11.0.yaml

@@ -32,14 +32,7 @@
                           BatchEncoding, BatchFeature)
 from transformers.models.auto.auto_factory import _BaseAutoModelClass
 from vllm import LLM, SamplingParams
-
-from vllm_ascend.utils import vllm_version_is
-
-if vllm_version_is("0.10.2"):
-    from vllm.config import TaskOption, _get_and_verify_dtype
-else:
-    from vllm.config.model import TaskOption, _get_and_verify_dtype
-
+from vllm.config.model import TaskOption, _get_and_verify_dtype
 from vllm.inputs import TextPrompt
 from vllm.outputs import RequestOutput
 from vllm.transformers_utils.utils import maybe_model_redirect

tests/e2e/conftest.py

@@ -19,12 +19,7 @@
 
 from typing import Dict, List, Optional, Sequence, Tuple, Union
 
-from vllm_ascend.utils import vllm_version_is
-
-if vllm_version_is("0.10.2"):
-    from vllm.sequence import PromptLogprobs, SampleLogprobs
-else:
-    from vllm.logprobs import PromptLogprobs, SampleLogprobs
+from vllm.logprobs import PromptLogprobs, SampleLogprobs
 
 TokensText = Tuple[List[int], str]
 

tests/e2e/model_utils.py

@@ -22,15 +22,8 @@
 import jsonschema
 import pytest
 import regex as re
-
-from vllm_ascend.utils import vllm_version_is
-
-if vllm_version_is("0.10.2"):
-    from vllm.sampling_params import GuidedDecodingParams, SamplingParams
-else:
-    from vllm.sampling_params import SamplingParams, StructuredOutputsParams
-
 from vllm.outputs import RequestOutput
+from vllm.sampling_params import SamplingParams, StructuredOutputsParams
 
 from tests.e2e.conftest import VllmRunner
 
@@ -91,27 +84,16 @@ def sample_json_schema():
 def test_guided_json_completion(guided_decoding_backend: str,
                                 sample_json_schema):
     runner_kwargs: Dict[str, Any] = {}
-    if vllm_version_is("0.10.2"):
-        sampling_params = SamplingParams(
-            temperature=1.0,
-            max_tokens=500,
-            guided_decoding=GuidedDecodingParams(json=sample_json_schema))
-        runner_kwargs = {
-            "seed": 0,
-            "guided_decoding_backend": guided_decoding_backend,
-        }
-    else:
-        sampling_params = SamplingParams(
-            temperature=1.0,
-            max_tokens=500,
-            structured_outputs=StructuredOutputsParams(
-                json=sample_json_schema))
-        runner_kwargs = {
-            "seed": 0,
-            "structured_outputs_config": {
-                "backend": guided_decoding_backend
-            },
-        }
+    sampling_params = SamplingParams(
+        temperature=1.0,
+        max_tokens=500,
+        structured_outputs=StructuredOutputsParams(json=sample_json_schema))
+    runner_kwargs = {
+        "seed": 0,
+        "structured_outputs_config": {
+            "backend": guided_decoding_backend
+        },
+    }
     with VllmRunner(MODEL_NAME, **runner_kwargs) as vllm_model:
         prompts = [
             f"Give an example JSON for an employee profile "
@@ -141,26 +123,16 @@ def test_guided_regex(guided_decoding_backend: str, sample_regex):
     if guided_decoding_backend == "outlines":
         pytest.skip("Outlines doesn't support regex-based guided decoding.")
     runner_kwargs: Dict[str, Any] = {}
-    if vllm_version_is("0.10.2"):
-        sampling_params = SamplingParams(
-            temperature=0.8,
-            top_p=0.95,
-            guided_decoding=GuidedDecodingParams(regex=sample_regex))
-        runner_kwargs = {
-            "seed": 0,
-            "guided_decoding_backend": guided_decoding_backend,
-        }
-    else:
-        sampling_params = SamplingParams(
-            temperature=0.8,
-            top_p=0.95,
-            structured_outputs=StructuredOutputsParams(regex=sample_regex))
-        runner_kwargs = {
-            "seed": 0,
-            "structured_outputs_config": {
-                "backend": guided_decoding_backend
-            },
-        }
+    sampling_params = SamplingParams(
+        temperature=0.8,
+        top_p=0.95,
+        structured_outputs=StructuredOutputsParams(regex=sample_regex))
+    runner_kwargs = {
+        "seed": 0,
+        "structured_outputs_config": {
+            "backend": guided_decoding_backend
+        },
+    }
 
     with VllmRunner(MODEL_NAME, **runner_kwargs) as vllm_model:
         prompts = [

tests/e2e/singlecard/test_guided_decoding.py

@@ -8,7 +8,6 @@
     FusedMoEPrepareAndFinalizeWithAll2All,
     FusedMoEPrepareAndFinalizeWithAllGather, FusedMoEPrepareAndFinalizeWithMC2,
     FusedMoEPrepareAndFinalizeWithNaiveMulticast)
-from vllm_ascend.utils import vllm_version_is
 
 
 class TestFusedMoEPrepareAndFinalize(unittest.TestCase):
@@ -231,12 +230,8 @@ def test_naive_multicast_prepare_finalize(self, mock_get_forward_context,
                                               mock_get_dp_group):
         # Mock forward context with DP metadata
         mock_context = MagicMock()
-        if vllm_version_is("0.10.2"):
-            mock_context.dp_metadata.cu_tokens_across_dp_cpu = torch.tensor(
-                [2, 5, 7])
-        else:
-            mock_context.dp_metadata.cu_tokens_across_sp.return_value = torch.tensor(
-                [2, 5, 7])
+        mock_context.dp_metadata.cu_tokens_across_sp.return_value = torch.tensor(
+            [2, 5, 7])
         mock_get_forward_context.return_value = mock_context
 
         # Setup DP group mock

tests/ut/ops/test_fused_moe_prepare_and_finalize.py

@@ -28,7 +28,7 @@
                                        AscendUnquantizedFusedMoEMethod)
 from vllm_ascend.ops.moe.experts_selector import select_experts
 from vllm_ascend.ops.moe.moe_mlp import cumsum_group_list, unified_apply_mlp
-from vllm_ascend.utils import AscendSocVersion, adapt_patch, vllm_version_is
+from vllm_ascend.utils import AscendSocVersion, adapt_patch
 
 adapt_patch(True)
 
@@ -92,11 +92,7 @@ def mock_finalize(hidden_states, **kwargs):
         return hidden_states
 
     mock_moe_comm_method.finalize.side_effect = mock_finalize
-
-    if vllm_version_is("0.10.2"):
-        dp_metadata = MagicMock(cu_tokens_across_dp_cpu=[5, 10])
-    else:
-        dp_metadata = MagicMock(num_tokens_across_dp_cpu=[5, 5])
+    dp_metadata = MagicMock(num_tokens_across_dp_cpu=[5, 5])
     mock_forward_context_obj = MagicMock(moe_comm_method=mock_moe_comm_method,
                                          moe_comm_type=MoECommType.MC2,
                                          max_tokens_across_dp=10,

tests/ut/ops/test_fused_ops.py

@@ -27,7 +27,7 @@
 from vllm_ascend.torchair.ops.torchair_fused_moe import (
     TorchairAscendFusedMoE, TorchairAscendUnquantizedFusedMoEMethod)
 from vllm_ascend.utils import adapt_patch  # noqa E402
-from vllm_ascend.utils import AscendSocVersion, vllm_version_is
+from vllm_ascend.utils import AscendSocVersion
 
 adapt_patch(True)
 
@@ -54,10 +54,7 @@ def mock_dp_and_tp_group(mocker):
 @pytest.fixture
 def mock_dist_env(mocker: MockerFixture):
     # init dist env patch
-    if vllm_version_is("0.10.2"):
-        dp_metadata = MagicMock(cu_tokens_across_dp_cpu=[5, 10])
-    else:
-        dp_metadata = MagicMock(num_tokens_across_dp_cpu=[5, 5])
+    dp_metadata = MagicMock(num_tokens_across_dp_cpu=[5, 5])
 
     with patch('torch.distributed.get_rank', return_value=0), \
          patch('torch.distributed.get_world_size', return_value=4), \

tests/ut/torchair/ops/test_torchair_fused_moe.py

@@ -19,6 +19,5 @@
 import vllm_ascend.patch.platform.patch_common.patch_distributed  # noqa
 import vllm_ascend.patch.platform.patch_common.patch_mamba_config  # noqa
 import vllm_ascend.patch.platform.patch_common.patch_multimodal_merge  # noqa
-import vllm_ascend.patch.platform.patch_common.patch_transformers_utils  # noqa
 import vllm_ascend.patch.worker.patch_common.patch_attention_selector  # noqa
 import vllm_ascend.patch.worker.patch_common.patch_attentionspec  # noqa

vllm_ascend/patch/platform/patch_common/__init__.py

@@ -1,87 +1,10 @@
 import ast
 
 import vllm.envs as envs
-from transformers import PretrainedConfig
-from vllm.config import ModelConfig
 from vllm.config.speculative import SpeculativeConfig
 from vllm.logger import logger
 
 
-# mypy: ignore-errors
-@property
-def is_deepseek_mla(self: ModelConfig):
-    if not hasattr(self.hf_text_config, "model_type"):
-        return False
-    elif self.hf_text_config.model_type in \
-        ('deepseek_v2', 'deepseek_v3', 'deepseek_mtp',
-            'kimi_k2', 'longcat_flash', 'deepseek_v32'):
-        return self.hf_text_config.kv_lora_rank is not None
-    elif self.hf_text_config.model_type == 'eagle':
-        # if the model is an EAGLE module, check for the
-        # underlying architecture
-        return self.hf_text_config.model.model_type in \
-                ('deepseek_v2', 'deepseek_v3', 'deepseek_v32') \
-            and self.hf_text_config.kv_lora_rank is not None
-    return False
-
-
-@staticmethod
-def hf_config_override(hf_config: PretrainedConfig) -> PretrainedConfig:
-    if hf_config.model_type in ("deepseek_v3", "deepseek_v32"):
-        hf_config.model_type = "deepseek_mtp"
-    if hf_config.model_type == "deepseek_mtp":
-        n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
-        hf_config.update({
-            "n_predict": n_predict,
-            "architectures": ["DeepSeekMTPModel"]
-        })
-
-    if hf_config.architectures[0] == "MiMoForCausalLM":
-        hf_config.model_type = "mimo_mtp"
-        n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
-        hf_config.update({
-            "num_hidden_layers": 0,
-            "n_predict": n_predict,
-            "architectures": ["MiMoMTPModel"]
-        })
-
-    if hf_config.architectures[0] == "Glm4MoeForCausalLM":
-        hf_config.model_type = "glm4_moe_mtp"
-        n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
-        hf_config.update({
-            "num_hidden_layers": 0,
-            "n_predict": n_predict,
-            "architectures": ["Glm4MoeMTPModel"]
-        })
-
-    if hf_config.model_type == "ernie4_5_moe":
-        hf_config.model_type = "ernie_mtp"
-    if hf_config.model_type == "ernie_mtp":
-        n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
-        hf_config.update({
-            "n_predict": n_predict,
-            "architectures": ["ErnieMTPModel"]
-        })
-
-    if hf_config.model_type == "qwen3_next":
-        hf_config.model_type = "qwen3_next_mtp"
-    if hf_config.model_type == "qwen3_next_mtp":
-        n_predict = getattr(hf_config, "num_nextn_predict_layers", None)
-        hf_config.update({
-            "n_predict": n_predict,
-            "architectures": ["Qwen3NextMTP"]
-        })
-    if hf_config.model_type == "longcat_flash":
-        hf_config.model_type = "longcat_flash_mtp"
-        n_predict = getattr(hf_config, "num_nextn_predict_layers", 1)
-        hf_config.update({
-            "n_predict": n_predict,
-            "architectures": ["LongCatFlashMTPModel"]
-        })
-
-    return hf_config
-
-
 def __post_init__(self):
 
     # Note: "method" is a new parameter that helps to extend the
@@ -308,6 +231,4 @@ def __post_init__(self):
                     self.draft_tensor_parallel_size))
 
 
-ModelConfig.is_deepseek_mla = is_deepseek_mla
 SpeculativeConfig.__post_init__ = __post_init__
-SpeculativeConfig.hf_config_override = hf_config_override