[Main2Main] Upgrade vllm commit to v0.15.0rc0

.github/workflows/_e2e_nightly_multi_node.yaml

@@ -32,7 +32,7 @@ on:
         description: how many pods will be pulled up via lws.yaml, indicates number of nodes we need
       vllm_version:
         required: false
-        default: "v0.14.1"
+        default: "v0.15.0rc0"
         type: string
         description: vllm version to use
       vllm_ascend_remote_url:

.github/workflows/bot_pr_create.yaml

@@ -37,7 +37,7 @@ jobs:
     steps:
       - name: Get vLLM version
         run: |
-          VLLM_COMMIT=dc917cceb877dfd13f98c538c4c96158047d98bd
+          VLLM_COMMIT=cf1167e50b809f18efd21fb3418dd75d2805b14f
           echo "VLLM_COMMIT=https://github.com/vllm-project/vllm/commit/$VLLM_COMMIT" >> $GITHUB_ENV
 
       - name: Checkout repository

.github/workflows/dockerfiles/Dockerfile.lint

@@ -27,7 +27,7 @@ RUN apt-get update -y && \
 
 ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
 # For lint purpose, actually we need make a main2main matching.
-ARG VLLM_COMMIT=dc917cceb877dfd13f98c538c4c96158047d98bd
+ARG VLLM_COMMIT=cf1167e50b809f18efd21fb3418dd75d2805b14f
 RUN git clone $VLLM_REPO /vllm-workspace/vllm && \
     cd /vllm-workspace/vllm && \
     git checkout $VLLM_COMMIT

.github/workflows/pr_test_full.yaml

@@ -75,7 +75,7 @@ jobs:
     name: e2e-full
     strategy:
       matrix:
-        vllm_version: [dc917cceb877dfd13f98c538c4c96158047d98bd, v0.14.1]
+        vllm_version: [cf1167e50b809f18efd21fb3418dd75d2805b14f, v0.15.0rc0]
     needs: [changes]
     if: ${{ needs.changes.outputs.e2e_tracker == 'true' || needs.changes.outputs.e2e_tracker == true }}
     uses: ./.github/workflows/_e2e_test.yaml

.github/workflows/pr_test_light.yaml

@@ -41,7 +41,7 @@ jobs:
   lint:
     uses: ./.github/workflows/_pre_commit.yml
     with:
-      vllm: dc917cceb877dfd13f98c538c4c96158047d98bd
+      vllm: cf1167e50b809f18efd21fb3418dd75d2805b14f
   changes:
     runs-on: linux-aarch64-a2-0
     outputs:
@@ -87,7 +87,7 @@ jobs:
     if: ${{ needs.lint.result == 'success' && (needs.changes.outputs.e2e_tracker == 'true' || needs.changes.outputs.ut_tracker == 'true') }}
     strategy:
       matrix:
-        vllm_version: [dc917cceb877dfd13f98c538c4c96158047d98bd, v0.14.1]
+        vllm_version: [cf1167e50b809f18efd21fb3418dd75d2805b14f, v0.15.0rc0]
     uses: ./.github/workflows/_unit_test.yaml
     with:
       vllm: ${{ matrix.vllm_version }}
@@ -99,7 +99,7 @@ jobs:
     name: e2e-light
     strategy:
       matrix:
-        vllm_version: [dc917cceb877dfd13f98c538c4c96158047d98bd, v0.14.1]
+        vllm_version: [cf1167e50b809f18efd21fb3418dd75d2805b14f, v0.15.0rc0]
     # Note (yikun): If CI resource are limited we can split job into two chain jobs
     needs: [lint, changes]
     # only trigger e2e test after lint passed and the change is e2e related with pull request.

.github/workflows/schedule_codecov_refresh.yaml

@@ -33,7 +33,7 @@ jobs:
     name: refresh codecov
     strategy:
       matrix:
-        vllm_version: [dc917cceb877dfd13f98c538c4c96158047d98bd]
+        vllm_version: [cf1167e50b809f18efd21fb3418dd75d2805b14f]
     uses: ./.github/workflows/_unit_test.yaml
     with:
       vllm: ${{ matrix.vllm_version }}

.github/workflows/schedule_nightly_test_a2.yaml

@@ -133,7 +133,7 @@ jobs:
               - Qwen3-Omni-30B-A3B-Instruct
     uses: ./.github/workflows/_e2e_nightly_single_node_models.yaml
     with:
-      vllm: v0.14.1
+      vllm: v0.15.0rc0
       runner: ${{ matrix.test_config.os }}
       model_list: ${{ toJson(matrix.test_config.model_list) }}
       image: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-910b-ubuntu22.04-py3.11'

.github/workflows/schedule_test_benchmarks.yaml

@@ -51,7 +51,7 @@ jobs:
     strategy:
       matrix:
         include:
-          - vllm_branch: v0.14.1
+          - vllm_branch: v0.15.0rc0
             vllm_ascend_branch: main
       max-parallel: 1
     container:

Dockerfile

@@ -48,7 +48,7 @@ RUN pip config set global.index-url ${PIP_INDEX_URL}
 
 # Install vLLM
 ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
-ARG VLLM_TAG=v0.14.1
+ARG VLLM_TAG=v0.15.0rc0
 RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
 # In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
 RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \

Dockerfile.310p

@@ -40,7 +40,7 @@ RUN pip config set global.index-url ${PIP_INDEX_URL}
 
 # Install vLLM
 ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
-ARG VLLM_TAG=v0.14.1
+ARG VLLM_TAG=v0.15.0rc0
 RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
 # In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
 RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \

Dockerfile.310p.openEuler

@@ -36,7 +36,7 @@ COPY . /vllm-workspace/vllm-ascend/
 
 # Install vLLM
 ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
-ARG VLLM_TAG=v0.14.1
+ARG VLLM_TAG=v0.15.0rc0
 RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
 # In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
 RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \

Dockerfile.a3

@@ -47,7 +47,7 @@ RUN apt-get update -y && \
 
 # Install vLLM
 ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
-ARG VLLM_TAG=v0.14.1
+ARG VLLM_TAG=v0.15.0rc0
 RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
 # In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
 RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \

Dockerfile.a3.openEuler

@@ -50,7 +50,7 @@ RUN yum update -y && \
 
 # Install vLLM
 ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
-ARG VLLM_TAG=v0.14.1
+ARG VLLM_TAG=v0.15.0rc0
 RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
 # In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
 RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \

Dockerfile.openEuler

@@ -50,7 +50,7 @@ RUN yum update -y && \
 
 # Install vLLM
 ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
-ARG VLLM_TAG=v0.14.1
+ARG VLLM_TAG=v0.15.0rc0
 RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
 # In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
 RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \

docs/source/community/versioning_policy.md

@@ -55,7 +55,7 @@ For main branch of vLLM Ascend, we usually make it compatible with the latest vL
 
 | vLLM Ascend | vLLM         | Python           | Stable CANN | PyTorch/torch_npu  |
 |-------------|--------------|------------------|-------------|--------------------|
-|     main    | dc917cceb877dfd13f98c538c4c96158047d98bd, v0.14.1 tag | >= 3.10, < 3.12   | 8.5.0 | 2.9.0 / 2.9.0 |
+|     main    | cf1167e50b809f18efd21fb3418dd75d2805b14f, v0.15.0rc0 tag | >= 3.10, < 3.12   | 8.5.0 | 2.9.0 / 2.9.0 |
 
 ## Release cadence
 

docs/source/conf.py

@@ -65,7 +65,7 @@
     # the branch of vllm, used in vllm clone
     # - main branch: 'main'
     # - vX.Y.Z branch: 'vX.Y.Z'
-    "vllm_version": "v0.14.1",
+    "vllm_version": "v0.15.0rc0",
     # the branch of vllm-ascend, used in vllm-ascend clone and image tag
     # - main branch: 'main'
     # - vX.Y.Z branch: latest vllm-ascend release tag
@@ -77,7 +77,7 @@
     # CANN image tag
     "cann_image_tag": "8.5.0-910b-ubuntu22.04-py3.11",
     # vllm version in ci
-    "ci_vllm_version": "v0.14.1",
+    "ci_vllm_version": "v0.15.0rc0",
 }
 
 # For cross-file header anchors

vllm_ascend/ascend_forward_context.py

@@ -19,7 +19,6 @@
     is_drafter_moe_model,
     is_moe_model,
     speculative_enable_dispatch_gmm_combine_decode,
-    vllm_version_is,
 )
 
 
@@ -59,8 +58,7 @@ def set_ascend_forward_context(
         "batch_descriptor": batch_descriptor,
     }
 
-    if not vllm_version_is("0.14.1"):
-        forward_context_kwargs["skip_compiled"] = skip_compiled
+    forward_context_kwargs["skip_compiled"] = skip_compiled
 
     with set_forward_context(**forward_context_kwargs):
         forward_context = get_forward_context()

vllm_ascend/attention/attention_v1.py

@@ -66,6 +66,8 @@
 class AscendAttentionBackend(AttentionBackend):
     accept_output_buffer: bool = True
 
+    forward_includes_kv_cache_update: bool = True
+
     @staticmethod
     def get_name() -> str:
         # HACK(Ronald1995): vllm `initialize_kv_cache` method in model runner v2 make
@@ -868,6 +870,9 @@ def reshape_and_cache(
                 self.key_cache, self.value_cache = kv_cache[0], kv_cache[1]
             slots = attn_metadata.slot_mapping
             encoder_decoder = self.attn_type == AttentionType.ENCODER_DECODER
+
+            slots = slots.to(torch.int32)
+
             DeviceOperator.reshape_and_cache(
                 key=key[: attn_metadata.num_actual_tokens] if not encoder_decoder else key,
                 value=value[: attn_metadata.num_actual_tokens] if not encoder_decoder else value,
@@ -932,8 +937,10 @@ def forward(
         num_tokens = query.shape[0]
         if attn_metadata is None:
             return output.fill_(0)
+
         if key is not None and value is not None:
             key, value = self.reshape_and_cache(key, value, kv_cache, attn_metadata)
+
         # pooling model branch
         if attn_metadata.model_runner_type == "pooling":
             attn_output = self._forward_encoder_attention(query, key, value, attn_metadata, output)

vllm_ascend/attention/mla_v1.py

@@ -44,17 +44,14 @@
 from vllm_ascend.ops.rotary_embedding import get_cos_and_sin_mla
 from vllm_ascend.ops.weight_prefetch import maybe_npu_prefetch
 from vllm_ascend.quantization.methods import AscendW8A8LinearMethod
-from vllm_ascend.utils import ACL_FORMAT_FRACTAL_ND, maybe_trans_nz, vllm_version_is, weak_ref_tensors
+from vllm_ascend.utils import ACL_FORMAT_FRACTAL_ND, maybe_trans_nz, weak_ref_tensors
 from vllm_ascend.worker.npu_input_batch import NPUInputBatch
 
 if TYPE_CHECKING:
     from vllm.v1.core.sched.output import SchedulerOutput
 
 # isort: off
-if vllm_version_is("0.14.1"):
-    from vllm.v1.attention.backends.mla.common import MLACommonMetadataBuilder  # type: ignore
-else:
-    from vllm.model_executor.layers.attention.mla_attention import MLACommonMetadataBuilder
+from vllm.model_executor.layers.attention.mla_attention import MLACommonMetadataBuilder
 # isort: on
 
 MAX_O_PROJ_PREFETCH_SIZE = 16 * 1024 * 1024

vllm_ascend/attention/sfa_v1.py

@@ -45,16 +45,13 @@
     enable_dsa_cp,
     enable_dsa_cp_with_layer_shard,
     maybe_trans_nz,
-    vllm_version_is,
 )
 from vllm_ascend.worker.npu_input_batch import NPUInputBatch
 
 if TYPE_CHECKING:
     from vllm.v1.core.sched.output import SchedulerOutput
-if vllm_version_is("0.14.1"):
-    from vllm.v1.attention.backends.mla.common import MLACommonMetadataBuilder  # type: ignore
-else:
-    from vllm.model_executor.layers.attention.mla_attention import MLACommonMetadataBuilder
+from vllm.model_executor.layers.attention.mla_attention import MLACommonMetadataBuilder
+
 # isort: on
 
 # token count limits within bmm_transpose operator

vllm_ascend/ops/fused_moe/fused_moe.py

@@ -512,6 +512,15 @@ def forward(
         hidden_states: torch.Tensor,
         router_logits: torch.Tensor,
     ) -> tuple[torch.Tensor, torch.Tensor]:
+        if self._shared_experts is None:
+            fused_out = AscendFusedMoE.forward(
+                self,
+                hidden_states=hidden_states,
+                router_logits=router_logits,
+            )
+            shared_out = None
+            return shared_out, fused_out
+
         shared_out, fused_out = AscendFusedMoE.forward(
             self,
             hidden_states=hidden_states,
@@ -571,6 +580,9 @@ def forward_impl(  # type: ignore[override]
         )
         routed_out = fused_moe_results.routed_out
 
+        if self._shared_experts is None:
+            return routed_out
+
         if self.multistream_overlap_gate:
             fc3_context = get_flash_common3_context()
             assert fc3_context is not None

vllm_ascend/ops/mm_encoder_attention.py

@@ -38,7 +38,6 @@ def __init__(
         scale: float | None = None,
         num_kv_heads: int | None = None,
         prefix: str = "",
-        multimodal_config: MultiModalConfig | None = None,
     ) -> None:
         """
         Args:
@@ -48,15 +47,13 @@ def __init__(
             num_kv_heads: number of kv heads.
             prefix: This has no effect, it is only here to make it easier to
                     swap between Attention and MMEncoderAttention.
-            multimodal_config: configs for multi-modal.
         """
         super().__init__(
             num_heads=num_heads,
             head_size=head_size,
             scale=scale,
             num_kv_heads=num_kv_heads,
             prefix=prefix,
-            multimodal_config=multimodal_config,
         )
 
     def reshape_qkv_to_3d(
@@ -84,13 +81,12 @@ def reshape_qkv_to_3d(
         return query, key, value
 
     def forward_oot(
-            self,
-            query: torch.Tensor,
-            key: torch.Tensor,
-            value: torch.Tensor,
-            cu_seqlens: torch.Tensor | None = None,
-            max_seqlen: torch.Tensor
-        | None = None,  # Only used for Flash Attention
+        self,
+        query: torch.Tensor,
+        key: torch.Tensor,
+        value: torch.Tensor,
+        cu_seqlens: torch.Tensor | None = None,
+        max_seqlen: torch.Tensor | None = None,  # Only used for Flash Attention
     ):
         bsz, q_len = query.size()[:2]
         kv_len = key.size(1)

vllm_ascend/patch/platform/patch_multiproc_executor.py

@@ -19,8 +19,6 @@
     set_multiprocessing_worker_envs,
 )
 
-from vllm_ascend.utils import vllm_version_is
-
 
 class AscendMultiprocExecutor(MultiprocExecutor):
     def _init_executor(self) -> None:
@@ -178,8 +176,7 @@ def make_worker_process(
             "death_pipe": death_reader,
             "shared_worker_lock": shared_worker_lock,
         }
-        if not vllm_version_is("0.14.1"):
-            process_kwargs["is_driver_worker"] = is_driver_worker
+        process_kwargs["is_driver_worker"] = is_driver_worker
         # Run EngineCore busy loop in background process.
         proc = context.Process(
             target=WorkerProc.worker_main,