[Feat][BugFix]Support the Qwen3-Next-80B-A3B-Instruct quantization model&Fix the NZ issue (#4245)

### What this PR does / why we need it?
Support the Qwen3-Next-80B-A3B-Instruct quantization model and Fix the
NZ issue. Triton kernel doesn't support data format nz, thus we skip
converting weight to nz on layer `conv1d`

- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

---------

Signed-off-by: IncSec <1790766300@qq.com>

vllm_ascend/ops/linear.py

@@ -45,7 +45,8 @@ class AscendUnquantizedLinearMethod(UnquantizedLinearMethod):
 
     def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
         super().process_weights_after_loading(layer)
-        if (is_enable_nz() and layer.weight.data.dtype
+        if "conv1d" not in layer.prefix and (
+                is_enable_nz() and layer.weight.data.dtype
                 in [torch.float16, torch.bfloat16]):
             layer.weight.data = torch_npu.npu_format_cast(
                 layer.weight.data, ACL_FORMAT_FRACTAL_NZ)

vllm_ascend/quantization/quant_config.py

@@ -222,6 +222,8 @@ packed_modules_model_mapping = {
         ],
         "gate_up_proj": ["gate_proj", "up_proj"],
         "in_proj": ["in_proj_qkvz", "in_proj_ba"],
+        "experts":
+        ["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"]
     },
     "qwen2_5_vl": {
         "qkv_proj": [

vllm_ascend/utils.py

@@ -59,7 +59,6 @@ _MIN_DP_BUFFER_SIZE = 50
 _IS_MOE_MODEL = None
 _ENABLE_SP = None
 _HAS_LAYER_IDX = None
-_ENABLE_NZ = None
 _SUBSCRIBED_COMPUTE_STREAMS = set()
 _GRAPH_PRINT_STREAM = None
 _GRAPH_PRINT_STREAM_LOCK = Lock()
@@ -129,14 +128,8 @@ def is_310p():
     return _IS_310P
 
 
-def is_enable_nz(vllm_config: Optional[VllmConfig] = None) -> bool:
-    global _ENABLE_NZ
-    if _ENABLE_NZ is None:
-        if not vllm_config:
-            raise ValueError(
-                "vllm_config must be provided when _ENABLE_NZ is None")
-        _ENABLE_NZ = envs_ascend.VLLM_ASCEND_ENABLE_NZ and vllm_config.model_config.hf_config.model_type != "qwen3_next"
-    return _ENABLE_NZ
+def is_enable_nz():
+    return envs_ascend.VLLM_ASCEND_ENABLE_NZ
 
 
 def sleep_mode_enabled():

vllm_ascend/worker/worker_v1.py

@@ -87,7 +87,6 @@ class NPUWorker(WorkerBase):
         # register patch for vllm
         from vllm_ascend.utils import adapt_patch
         adapt_patch()
-        is_enable_nz(vllm_config)
         # Register ops when worker init.
         from vllm_ascend import ops
         ops.register_dummy_fusion_op()