[Refactor] refactor patch module (#3555)

### What this PR does / why we need it?
we notice that `patch_main` is never used. Usually the patch is for all
version. And if it's for specified version, we can use `vllm_version_is`
instead. So let's remove the useless sub folder in patch module to make
it clear.


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

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

docs/source/developer_guide/feature_guide/patch.md

@@ -20,13 +20,9 @@ In `vllm_ascend/patch`, you can see the code structure as follows:
 vllm_ascend
 ├── patch
 │   ├── platform
-│   │   ├── patch_0_9_2
-│   │   ├── patch_common
-│   │   ├── patch_main
+│   │   ├── patch_xxx.py
 │   ├── worker
-│   │   ├── patch_0_9_2
-│   │   ├── patch_common
-│   │   ├── patch_main
+│   │   ├── patch_yyy.py
 └───────────
 ```
 
@@ -36,19 +32,13 @@ vllm_ascend
 - **worker**: The patch code in this directory is for patching the code in vLLM worker process. It's called by `vllm_ascend/worker/worker_v1::NPUWorker::__init__` when the vLLM worker process is initialized.
   - For both online and offline mode, vLLM engine core process calls the worker patch here `vllm/vllm/worker/worker_base.py::WorkerWrapperBase.init_worker` when initializing the worker process.
 
-In both **platform** and **worker** folder, there are several patch modules. They are used for patching different version of vLLM.
-
-- `patch_0_10_0`: This module is used for patching vLLM 0.10.0. The version is always the nearest version of vLLM. Once vLLM is released, we will drop this patch module and bump to a new version. For example, `patch_0_10_0` is used for patching vLLM 0.10.0.
-- `patch_main`: This module is used for patching the code in vLLM main branch.
-- `patch_common`: This module is used for patching both vLLM 0.10.0 and vLLM main branch.
-
 ## How to write a patch
 
 Before writing a patch, following the principle above, we should patch the least code. If it's necessary, we can patch the code in either **platform** and **worker** folder. Here is an example to patch `distributed` module in vLLM.
 
 1. Decide which version of vLLM we should patch. For example, after analysis, here we want to patch both 0.10.0 and main of vLLM.
 2. Decide which process we should patch. For example, here `distributed` belongs to the vLLM main process, so we should patch `platform`.
-3. Create the patch file in the right folder. The file should be named as `patch_{module_name}.py`. The example here is `vllm_ascend/patch/platform/patch_common/patch_distributed.py`.
+3. Create the patch file in the right folder. The file should be named as `patch_{module_name}.py`. The example here is `vllm_ascend/patch/platform/patch_distributed.py`.
 4. Write your patch code in the new file. Here is an example:
 
     ```python
@@ -61,7 +51,7 @@ Before writing a patch, following the principle above, we should patch the least
     vllm.distributed.parallel_state.destroy_model_parallel = patch_destroy_model_parallel
     ```
 
-5. Import the patch file in `__init__.py`. In this example, add `import vllm_ascend.patch.platform.patch_common.patch_distributed` into `vllm_ascend/patch/platform/patch_common/__init__.py`.
+5. Import the patch file in `__init__.py`. In this example, add `import vllm_ascend.patch.platform.patch_distributed` into `vllm_ascend/patch/platform/__init__.py`.
 6. Add the description of the patch in `vllm_ascend/patch/__init__.py`. The description format is as follows:
 
     ```

tests/ut/patch/worker/patch_common/test_patch_distributed.py

@@ -19,8 +19,7 @@ import torch
 from vllm.distributed.parallel_state import GroupCoordinator
 
 from tests.ut.base import TestBase
-from vllm_ascend.patch.worker.patch_common.patch_distributed import \
-    GroupCoordinatorPatch
+from vllm_ascend.patch.worker.patch_distributed import GroupCoordinatorPatch
 
 
 class TestPatchDistributed(TestBase):

tests/ut/patch/worker/patch_common/test_patch_minicpm.py

@@ -18,7 +18,7 @@ from unittest.mock import MagicMock
 import torch
 
 from tests.ut.base import TestBase
-from vllm_ascend.patch.worker.patch_common.patch_minicpm import forward
+from vllm_ascend.patch.worker.patch_minicpm import forward
 
 
 class TestPatchMiniCPM(TestBase):

vllm_ascend/patch/__init__.py

@@ -23,11 +23,6 @@
 #           `vllm_ascend.utils.adapt_patch(is_global_patch=False)` in
 #           each worker's `__init__` function.
 #
-# Then in each kind of patch, there are three folders:
-# - patch_0_10_0: contains the patches applied when vllm version is 0.10.0.
-# - patch_main: contains the patches applied when vllm version is main branch.
-# - patch_common: contains the patches applied in both 0.10.0 and main branch.
-#
 # Once a new patch is added in vllm-ascend, please add the patch description into this file as well.
 # ----------------------------------------------------------------------------------
 
@@ -35,7 +30,7 @@
 # --------------------------------
 # * Platform Patch:
 # =================
-# ** File: platform/patch_common/patch_distributed.py**
+# ** File: platform/patch_distributed.py**
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 #   1. `vllm.config.ParallelConfig.get_next_dp_init_port`
 #    Why:
@@ -56,7 +51,7 @@
 #    Future Plan:
 #       Find a better way to support tensor alignment for 310p without this patch.
 #
-# ** File: worker/patch_common/patch_multimodal_merge.py**
+# ** File: worker/patch_multimodal_merge.py**
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 #   1. `vllm.model_executor.models.utils._merge_multimodal_embeddings`
 #    Why:
@@ -70,7 +65,7 @@
 #
 # * Worker Patch:
 # ===============
-# ** File: worker/patch_common/patch_minicpm.py **
+# ** File: worker/patch_minicpm.py **
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 #   1. `vllm.model_executor.models.minicpm.MiniCPMAttention.forward`
 #    Why:
@@ -84,7 +79,7 @@
 #    Future Plan:
 #       Keep this patch in vllm-ascend.
 #
-# ** File: worker/patch_common/patch_distributed.py **
+# ** File: worker/patch_distributed.py **
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 #   1. `vllm.distributed.parallel_state.GroupCoordinator`
 #   (1) __init__()
@@ -120,7 +115,7 @@
 #       - https://github.com/vllm-project/vllm/pull/21591
 #    Future Plan:
 #       Revert it when vLLM merge #21591 and release new version
-# ** File: worker/patch_common/patch_logits.py **
+# ** File: worker/patch_logits.py **
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 #   1. `vllm._custom_ops.apply_repetition_penalties`
 #    Why:

vllm_ascend/patch/platform/__init__.py

@@ -14,5 +14,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from vllm_ascend.patch.platform import patch_common  # noqa: F401
-from vllm_ascend.patch.platform import patch_main  # noqa: F401
+import os
+
+import vllm_ascend.patch.platform.patch_config  # noqa
+import vllm_ascend.patch.platform.patch_distributed  # noqa
+import vllm_ascend.patch.platform.patch_mamba_config  # noqa
+
+if os.getenv("DYNAMIC_EPLB", False) or os.getenv("EXPERT_MAP_RECORD", False):
+    import vllm_ascend.patch.platform.patch_multiproc_executor  # noqa

vllm_ascend/patch/platform/patch_common/__init__.py

@@ -1,40 +0,0 @@
-#
-# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
-# This file is a part of the vllm-ascend project.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-import os
-
-from vllm.logger import logger
-
-import vllm_ascend.patch.platform.patch_common.patch_config  # noqa
-import vllm_ascend.patch.platform.patch_common.patch_distributed  # noqa
-import vllm_ascend.patch.platform.patch_common.patch_mamba_config  # noqa
-
-
-def patch_v1_executor():
-    try:
-        dynamic_eplb = os.getenv("DYNAMIC_EPLB", False) or os.getenv(
-            "EXPERT_MAP_RECORD", False)
-        if dynamic_eplb:
-            import vllm_ascend.patch.platform.patch_common.patch_multiproc_executor  # noqa
-        else:
-            logger.warning("Do not patch v1 executor.")
-    except RuntimeError as e:
-        logger.warning(
-            f"Fail to patch v1 executor, please add environment params DYNAMIC_EPLB or EXPERT_MAP_RECORD : {e}"
-        )
-
-
-patch_v1_executor()

vllm_ascend/patch/platform/patch_main/__init__.py

@@ -1,16 +0,0 @@
-#
-# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
-# This file is a part of the vllm-ascend project.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#

vllm_ascend/patch/worker/__init__.py

@@ -15,5 +15,15 @@
 # limitations under the License.
 #
 
-from vllm_ascend.patch.worker import patch_common  # noqa: F401
-from vllm_ascend.patch.worker import patch_main  # noqa: F401
+from vllm.triton_utils import HAS_TRITON
+
+if HAS_TRITON:
+    import vllm_ascend.patch.worker.patch_triton
+
+# isort: off
+import vllm_ascend.patch.worker.patch_distributed  # noqa
+import vllm_ascend.patch.worker.patch_logits  # noqa
+import vllm_ascend.patch.worker.patch_roberta  # noqa
+import vllm_ascend.patch.worker.patch_weight_loader  # noqa
+import vllm_ascend.patch.worker.patch_multimodal_merge  # noqa
+import vllm_ascend.patch.worker.patch_minicpm  # noqa

vllm_ascend/patch/worker/patch_common/__init__.py

@@ -1,29 +0,0 @@
-#
-# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
-# This file is a part of the vllm-ascend project.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-
-from vllm.triton_utils import HAS_TRITON
-
-if HAS_TRITON:
-    import vllm_ascend.patch.worker.patch_common.patch_triton
-
-# isort: off
-import vllm_ascend.patch.worker.patch_common.patch_distributed  # noqa
-import vllm_ascend.patch.worker.patch_common.patch_logits  # noqa
-import vllm_ascend.patch.worker.patch_common.patch_roberta  # noqa
-import vllm_ascend.patch.worker.patch_common.patch_weight_loader  # noqa
-import vllm_ascend.patch.worker.patch_common.patch_multimodal_merge  # noqa
-import vllm_ascend.patch.worker.patch_common.patch_minicpm  # noqa

vllm_ascend/patch/worker/patch_common/patch_attention_selector.py

@@ -1,100 +0,0 @@
-#
-# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
-# This file is a part of the vllm-ascend project.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-# mypy: ignore-errors
-from functools import cache
-from typing import Optional
-
-import torch
-import vllm
-import vllm.envs as envs
-from vllm.attention.backends.abstract import AttentionBackend
-from vllm.attention.selector import (backend_name_to_enum,
-                                     get_global_forced_attn_backend)
-from vllm.platforms import _Backend, current_platform
-from vllm.utils import resolve_obj_by_qualname
-
-
-def get_attn_backend(  # type: ignore[misc]
-    head_size: int,
-    dtype: torch.dtype,
-    kv_cache_dtype: Optional[str],
-    block_size: int,
-    use_mla: bool = False,
-    use_sfa: bool = False,
-    has_sink: bool = False,
-) -> type[AttentionBackend]:
-    """Selects which attention backend to use and lazily imports it."""
-    # Accessing envs.* behind an @lru_cache decorator can cause the wrong
-    # value to be returned from the cache if the value changes between calls.
-    # To avoid this, we read envs.VLLM_USE_V1 here and pass it explicitly to the
-    # private function.
-    return _cached_get_attn_backend(
-        head_size=head_size,
-        dtype=dtype,
-        kv_cache_dtype=kv_cache_dtype,
-        block_size=block_size,
-        use_v1=envs.VLLM_USE_V1,
-        use_mla=use_mla,
-        use_sfa=use_sfa,
-        has_sink=has_sink,
-    )
-
-
-@cache
-def _cached_get_attn_backend(
-    head_size: int,
-    dtype: torch.dtype,
-    kv_cache_dtype: Optional[str],
-    block_size: int,
-    use_v1: bool = False,
-    use_mla: bool = False,
-    use_sfa: bool = False,
-    has_sink: bool = False,
-    use_sparse: bool = False,
-) -> type[AttentionBackend]:
-    # Check whether a particular choice of backend was
-    # previously forced.
-    #
-    # THIS SELECTION OVERRIDES THE VLLM_ATTENTION_BACKEND
-    # ENVIRONMENT VARIABLE.
-    selected_backend = None
-    backend_by_global_setting: Optional[_Backend] = (
-        get_global_forced_attn_backend())
-    if backend_by_global_setting is not None:
-        selected_backend = backend_by_global_setting
-    else:
-        # Check the environment variable and override if specified
-        backend_by_env_var: Optional[str] = envs.VLLM_ATTENTION_BACKEND
-        if backend_by_env_var is not None:
-            selected_backend = backend_name_to_enum(backend_by_env_var)
-            if selected_backend is None:
-                raise ValueError(
-                    f"Invalid attention backend: '{backend_by_env_var}'. "
-                    f"Valid backends are: {list(_Backend.__members__.keys())}")
-
-    # get device-specific attn_backend
-    attention_cls = current_platform.get_attn_backend_cls(
-        selected_backend, head_size, dtype, kv_cache_dtype, block_size, use_v1,
-        use_mla, use_sfa, has_sink)
-    if not attention_cls:
-        raise ValueError(
-            f"Invalid attention backend for {current_platform.device_name}")
-    return resolve_obj_by_qualname(attention_cls)
-
-
-vllm.attention.get_attn_backend = get_attn_backend
-vllm.attention.selector._cached_get_attn_backend = _cached_get_attn_backend

vllm_ascend/patch/worker/patch_main/__init__.py

@@ -1,16 +0,0 @@
-#
-# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
-# This file is a part of the vllm-ascend project.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#

vllm_ascend/patch/worker/patch_triton.py

@@ -13,4 +13,4 @@ vllm.model_executor.layers.mamba.ops.causal_conv1d.causal_conv1d_update = causal
 vllm.model_executor.layers.mamba.ops.causal_conv1d.causal_conv1d_fn = causal_conv1d_fn
 vllm.model_executor.layers.fla.ops.fused_recurrent.fused_recurrent_gated_delta_rule_fwd_kernel = fused_recurrent_gated_delta_rule_fwd_kernel
 vllm.model_executor.layers.fla.ops.layernorm_guard.LayerNormFn = LayerNormFn
-vllm.model_executor.layers.fla.ops.chunk.chunk_gated_delta_rule = torch_chunk_gated_delta_rule
+vllm.model_executor.layers.fla.ops.chunk.chunk_gated_delta_rule = torch_chunk_gated_delta_rule

vllm_ascend/torchair/torchair_model_runner.py

@@ -390,7 +390,7 @@ class NPUTorchairModelRunner(NPUModelRunner):
         if is_310p():
             # on 300I Duo platform, we need to patch broadcast. however, this patch will be
             # overwritten by patch_for_hcom in torchair. so we need to re-patch it here.
-            from vllm_ascend.patch.platform.patch_common.patch_distributed import \
+            from vllm_ascend.patch.platform.patch_distributed import \
                 communication_adaptation_310p
             communication_adaptation_310p()