Upgrade vllm commit hash to 1216

.github/workflows/pr_test_full.yaml

@@ -74,7 +74,7 @@ jobs:
     name: e2e-full
     strategy:
       matrix:
-        vllm_version: [4429d934de3c5cc327b0d7aec8e473aeba38db90, v0.12.0]
+        vllm_version: [releases/v0.13.0, v0.12.0]
     needs: [changes]
     if: ${{ needs.changes.outputs.e2e_tracker == 'true' }}
     uses: ./.github/workflows/_e2e_test.yaml

.github/workflows/pr_test_light.yaml

@@ -42,7 +42,7 @@ jobs:
   lint:
     uses: ./.github/workflows/_pre_commit.yml
     with:
-      vllm: 4429d934de3c5cc327b0d7aec8e473aeba38db90
+      vllm: releases/v0.13.0
   changes:
     runs-on: linux-aarch64-a2-0
     outputs:
@@ -90,7 +90,7 @@ jobs:
         SOC_VERSION: ascend910b1
     strategy:
       matrix:
-        vllm_version: [4429d934de3c5cc327b0d7aec8e473aeba38db90, v0.12.0]
+        vllm_version: [releases/v0.13.0, v0.12.0]
 
     steps:
       - name: Free up disk space
@@ -154,7 +154,7 @@ jobs:
     name: e2e-light
     strategy:
       matrix:
-        vllm_version: [4429d934de3c5cc327b0d7aec8e473aeba38db90, v0.12.0]
+        vllm_version: [releases/v0.13.0, v0.12.0]
     # 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.

docs/source/community/versioning_policy.md

@@ -50,7 +50,7 @@ If you're using v0.7.3, don't forget to install [mindie-turbo](https://pypi.org/
 For main branch of vLLM Ascend, we usually make it compatible with the latest vLLM release and a newer commit hash of vLLM. Please note that this table is usually updated. Please check it regularly.
 | vLLM Ascend | vLLM         | Python           | Stable CANN | PyTorch/torch_npu  |
 |-------------|--------------|------------------|-------------|--------------------|
-|     main    | 4429d934de3c5cc327b0d7aec8e473aeba38db90,  v0.12.0 tag | >= 3.10, < 3.12   | 8.3.RC2 | 2.8.0 / 2.8.0 |
+|     main    | releases/v0.13.0,  v0.12.0 tag | >= 3.10, < 3.12   | 8.3.RC2 | 2.8.0 / 2.8.0 |
 
 ## Release cadence
 

tests/e2e/multicard/test_offline_inference_distributed.py

@@ -113,11 +113,9 @@ def test_sp_for_qwen3_moe() -> None:
                     dtype="auto",
                     tensor_parallel_size=2,
                     distributed_executor_backend="mp",
-                    compilation_config={
-                        "pass_config": {
-                            "enable_sequence_parallelism": True
-                        }
-                    },
+                    compilation_config={"pass_config": {
+                        "enable_sp": True
+                    }},
                     enable_expert_parallel=True,
                     enforce_eager=True) as vllm_model:
         vllm_model.generate(example_prompts, sampling_params)

vllm_ascend/platform.py

@@ -355,23 +355,15 @@ def import_kernels(cls) -> None:
         CUSTOM_OP_REGISTERED = True
 
     @classmethod
-    def get_attn_backend_cls(
-        cls,
-        selected_backend,
-        head_size,
-        dtype,
-        kv_cache_dtype,
-        block_size,
-        use_mla,
-        has_sink=False,
-        use_sparse=False,
-        # NOTE: Please pay special attention to the order of these parameters.
-        # Although we are only using some of them so far
-        # vllm passes them in sequence when using this interface.
-        use_mm_prefix: bool = False,
-        attn_type: str | None = None,
-    ):
-        # choose attention backend based on use_mla
+    def get_attn_backend_cls(cls, selected_backend, *args, **kwargs):
+        if "attn_selector_config" in kwargs:
+            use_mla = kwargs["attn_selector_config"].use_mla
+            use_sparse = kwargs["attn_selector_config"].use_sparse
+        else:
+            use_mla = kwargs.get("use_mla",
+                                 args[4] if len(args) >= 5 else None)
+            use_sparse = kwargs.get("use_sparse",
+                                    args[6] if len(args) >= 7 else None)
         backend_map = {
             (True, False): "vllm_ascend.attention.mla_v1.AscendMLABackend",
             (False, False):