[Platform][BugFix] Preserve hybrid block size on Ascend (#7528)

### What this PR does / why we need it
This PR fixes a startup regression for Ascend hybrid attention + mamba
models after upgrading to vLLM `0.18.0`.
However, after the vLLM `0.18.0` upgrade, worker initialization still
calls the generic platform hook:
- `current_platform.update_block_size_for_backend(vllm_config)`

### How this PR fixes it

This PR keeps the fix strictly inside `vllm-ascend`.

It adds an Ascend override for
`NPUPlatform.update_block_size_for_backend()`:

- for hybrid models, do not run the generic upstream block-size fallback
- preserve the block size that was already computed by the hybrid
model-specific config logic
- for non-hybrid models, keep the original upstream behavior unchanged

- vLLM version: v0.18.0
- vLLM main:
8b6325758c
---------
Signed-off-by: maoxx241 <maomaoyu870@gmail.com>
Signed-off-by: Mengqing Cao <cmq0113@163.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>

tests/ut/test_platform.py

@@ -307,6 +307,26 @@ class TestNPUPlatform(TestBase):
 
         self.assertEqual(vllm_config.cache_config.block_size, 128)
 
+    def test_update_block_size_for_backend_preserves_hybrid_block_size(self):
+        vllm_config = TestNPUPlatform.mock_vllm_config()
+        vllm_config.model_config.is_hybrid = True
+        vllm_config.cache_config.block_size = 1024
+        vllm_config.cache_config.user_specified_block_size = False
+
+        self.platform.update_block_size_for_backend(vllm_config)
+
+        self.assertEqual(vllm_config.cache_config.block_size, 1024)
+
+    def test_update_block_size_for_backend_preserves_user_block_size(self):
+        vllm_config = TestNPUPlatform.mock_vllm_config()
+        vllm_config.model_config.is_hybrid = False
+        vllm_config.cache_config.block_size = 512
+        vllm_config.cache_config.user_specified_block_size = True
+
+        self.platform.update_block_size_for_backend(vllm_config)
+
+        self.assertEqual(vllm_config.cache_config.block_size, 512)
+
     @patch("vllm_ascend.quantization.utils.maybe_auto_detect_quantization")
     @patch("vllm_ascend.utils.get_ascend_device_type", return_value=AscendDeviceType.A3)
     @patch("vllm_ascend.ascend_config.init_ascend_config")

vllm_ascend/platform.py

@@ -172,6 +172,20 @@ class NPUPlatform(Platform):
     def inference_mode(cls):
         return torch.inference_mode()
 
+    @classmethod
+    def update_block_size_for_backend(cls, vllm_config: VllmConfig) -> None:
+        cache_config = vllm_config.cache_config
+        if cache_config.user_specified_block_size:
+            # User specified --block-size; keep it.
+            return
+        model_config = vllm_config.model_config
+        if model_config is not None and model_config.is_hybrid:
+            # Hybrid attention+mamba models rely on the model-specific sizing
+            # logic rather than the generic platform default.
+            return
+
+        super().update_block_size_for_backend(vllm_config)
+
     @classmethod
     def set_device(cls, device: torch.device):
         torch.npu.set_device(device)