[Model]Add Qwen3-Omni quantization Ascend NPU adaptation and optimization (#6828)

### What this PR does / why we need it?
This pull request is for quantization adaptation of Qwen3Omni, and it
achieves operator-level optimization and AUT (Auto-Quantization Tuning)
component optimization through patch-based modifications.
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.15.0
- vLLM main:
83b47f67b1

---------

Signed-off-by: tanhaoan333 <tanhaoan@huawei.com>

vllm_ascend/quantization/methods/w8a8_dynamic.py

@@ -79,6 +79,11 @@ class AscendW8A8DynamicLinearMethod(AscendLinearScheme):
         tp_rank: int | None = 0,
     ) -> torch.Tensor:
         quantized_x, pertoken_scale = torch_npu.npu_dynamic_quant(x)
+        need_unsqz = False
+        if pertoken_scale.dim() == 2:
+            need_unsqz = True
+            quantized_x = quantized_x.squeeze(dim=1)
+            pertoken_scale = pertoken_scale.squeeze(dim=1)
         output = torch_npu.npu_quant_matmul(
             quantized_x,
             layer.weight,
@@ -87,6 +92,8 @@ class AscendW8A8DynamicLinearMethod(AscendLinearScheme):
             bias=bias,
             output_dtype=x.dtype,
         )
+        if need_unsqz:
+            output = output.unsqueeze(dim=1)
         return output
 
     def process_weights_after_loading(self, layer):

vllm_ascend/quantization/modelslim_config.py

@@ -64,6 +64,13 @@ QUANT_MODEL_PREFIX_MAPPINGS: dict[str, dict[str, str]] = {
         "mm_projector.linear_1": "mm_projector.proj.0",
         "mm_projector.linear_2": "mm_projector.proj.2",
     },
+    "qwen3_omni_moe_thinker": {
+        "thinker.lm_head.": "language_model.lm_head.",
+        "thinker.model.": "language_model.model.",
+        "thinker.": "",
+        "lm_head.": "language_model.lm_head.",
+        "model.": "language_model.model.",
+    },
 }
 
 # key: model_type
@@ -186,6 +193,18 @@ packed_modules_model_mapping: dict[str, dict[str, list[str]]] = {
         ],
         "experts": ["experts.0.w1", "experts.0.w2", "experts.0.w3"],
     },
+    "qwen3_omni_moe_text": {
+        "qkv_proj": [
+            "q_proj",
+            "k_proj",
+            "v_proj",
+        ],
+        "gate_up_proj": [
+            "gate_proj",
+            "up_proj",
+        ],
+        "experts": ["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"],
+    },
 }
 
 
@@ -457,7 +476,10 @@ class AscendModelSlimConfig(QuantizationConfig):
                         "to have the same precision."
                     )
         else:
-            is_skipped = self.quant_description[prefix + ".weight"] == "FLOAT"
+            is_skipped = any(
+                key.startswith(prefix) and key.endswith(".weight") and value == "FLOAT"
+                for key, value in self.quant_description.items()
+            )
 
         assert is_skipped is not None
         return is_skipped