[BugFix] Fix Qwen3-Next because of vllm #24982 (#3221)

- Fixes Qwen3-Next because of vllm #24982

### Does this PR introduce _any_ user-facing change?
N/A

### How was this patch tested?
```
def main():
    prompts = [
        "窗前明月光，",
        "The president of the United States is Mr.",
        "The capital of France is",
        "The future of AI is",
        "感时花溅泪，",
        "家书抵万金啥意思？",
        "plz tell me a story: ",
    ]

    # Create a sampling params object.
    sampling_params = SamplingParams(max_tokens=100, temperature=0.6, top_k=40, top_p=0.95)
    # Create an LLM.
    llm = LLM(
        model="Qwen/Qwen3-Next-80B-A3B-Instruct",
              tensor_parallel_size=4,
              enforce_eager=True,
              trust_remote_code=True,
              max_model_len=256,
              gpu_memory_utilization=0.7,
              block_size=64
              )

    # Generate texts from the prompts.
    outputs = llm.generate(prompts, sampling_params)
    for output in outputs:
        prompt = output.prompt
        generated_text = output.outputs[0].text
        print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```


- vLLM version: v0.10.2
- vLLM main:
vllm-project/vllm@releases/v0.11.0

---------

Signed-off-by: Icey <[email protected]>

Author: wxsIcey

vllm_ascend/models/qwen3_next.py

@@ -51,13 +51,10 @@
 from vllm.transformers_utils.configs import Qwen3NextConfig
 from vllm.v1.attention.backends.gdn_attn import GDNAttentionMetadata
 
-from vllm.model_executor.models.qwen3_next import Qwen3NextAttention  # isort: skip
-from vllm.model_executor.models.qwen3_next import Qwen3NextDecoderLayer  # isort: skip
-from vllm.model_executor.models.qwen3_next import Qwen3NextForCausalLM  # isort: skip
-from vllm.model_executor.models.qwen3_next import Qwen3NextGatedDeltaNet  # isort: skip
-from vllm.model_executor.models.qwen3_next import Qwen3NextModel  # isort: skip
-from vllm.model_executor.models.qwen3_next import Qwen3NextSparseMoeBlock  # isort: skip
-from vllm.model_executor.models.qwen3_next import fused_gdn_gating  # isort: skip
+from vllm.model_executor.models.qwen3_next import (  # isort: skip
+    Qwen3NextAttention, Qwen3NextDecoderLayer, Qwen3NextForCausalLM,
+    Qwen3NextGatedDeltaNet, Qwen3NextModel, Qwen3NextSparseMoeBlock,
+    fused_gdn_gating)
 
 
 class CustomQwen3NextGatedDeltaNet(Qwen3NextGatedDeltaNet, MambaBase):
@@ -429,17 +426,16 @@ class CustomQwen3NextDecoderLayer(Qwen3NextDecoderLayer):
 
     def __init__(
         self,
-        config: Qwen3NextConfig,
+        vllm_config: VllmConfig,
         layer_type: str,
-        model_config: Optional[ModelConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
-        quant_config: Optional[QuantizationConfig] = None,
-        speculative_config: Optional[SpeculativeConfig] = None,
         prefix: str = "",
-        enable_eplb: bool = False,
     ) -> None:
         nn.Module.__init__(self)
-        self.config = config
+        config = vllm_config.model_config.hf_config
+        model_config = vllm_config.model_config
+        cache_config = vllm_config.cache_config
+        quant_config = vllm_config.quant_config
+        speculative_config = vllm_config.speculative_config
 
         self.layer_type = layer_type
         self.layer_idx = extract_layer_index(prefix)
@@ -468,12 +464,8 @@ def __init__(
         if (self.layer_idx not in mlp_only_layers) and (
                 config.num_experts > 0 and
             (self.layer_idx + 1) % config.decoder_sparse_step == 0):
-            self.mlp = Qwen3NextSparseMoeBlock(
-                config=config,
-                quant_config=quant_config,
-                prefix=f"{prefix}.mlp",
-                enable_eplb=enable_eplb,
-            )
+            self.mlp = Qwen3NextSparseMoeBlock(vllm_config=vllm_config,
+                                               prefix=f"{prefix}.mlp")
         else:
             self.mlp = Qwen3NextMLP(
                 hidden_size=config.hidden_size,
@@ -493,14 +485,14 @@ def __init__(
                 torch.zeros(
                     1,
                     1,
-                    self.config.hidden_size,
+                    config.hidden_size,
                     dtype=config.torch_dtype,
                 ), )
             self.ffn_layer_scale = torch.nn.Parameter(
                 torch.zeros(
                     1,
                     1,
-                    self.config.hidden_size,
+                    config.hidden_size,
                     dtype=config.torch_dtype,
                 ), )
 
@@ -511,13 +503,8 @@ class CustomQwen3NextModel(Qwen3NextModel):
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         nn.Module.__init__(self)
         config: Qwen3NextConfig = vllm_config.model_config.hf_config
-        model_config = vllm_config.model_config
-        cache_config = vllm_config.cache_config
-        quant_config = vllm_config.quant_config
         parallel_config = vllm_config.parallel_config
         lora_config = vllm_config.lora_config
-        speculative_config = vllm_config.speculative_config
-        enable_eplb = parallel_config.enable_eplb
         eplb_config = parallel_config.eplb_config
         self.num_redundant_experts = eplb_config.num_redundant_experts
 
@@ -534,14 +521,9 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
 
         def get_layer(prefix: str):
             return CustomQwen3NextDecoderLayer(
-                config,
+                vllm_config,
                 layer_type=config.layer_types[extract_layer_index(prefix)],
-                model_config=model_config,
-                cache_config=cache_config,
-                quant_config=quant_config,
-                speculative_config=speculative_config,
                 prefix=prefix,
-                enable_eplb=enable_eplb,
             )
 
         self.start_layer, self.end_layer, self.layers = make_layers(