remove unused code

vllm/compilation/decorators.py

@@ -221,8 +221,7 @@ def _support_torch_compile(
         # torch.compiler.is_compiling() means we are inside the compilation
         # e.g. TPU has the compilation logic in model runner, so we don't
         # need to compile the model inside.
-        skip_cuda_graphs = get_forward_context().skip_cuda_graphs
-        if envs.VLLM_ENABLE_TBO and skip_cuda_graphs:
+        if envs.VLLM_ENABLE_TBO and get_forward_context().skip_cuda_graphs:
             return self.forward(*args, **kwargs)
 
         if self.do_not_compile or torch.compiler.is_compiling() or get_profilling():

vllm/model_executor/layers/fused_moe/layer.py

@@ -437,7 +437,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
         logical_to_physical_map: Optional[torch.Tensor] = None,
         logical_replica_count: Optional[torch.Tensor] = None,
         use_nn_moe: Optional[bool] = False,
-        routed_scaling_factor: Optional[float] = None,
         use_fused_gate: Optional[bool] = False,
     ) -> torch.Tensor:
         if enable_eplb:
@@ -468,7 +467,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
             logical_to_physical_map=logical_to_physical_map,
             logical_replica_count=logical_replica_count,
             use_nn_moe=use_nn_moe,
-            routed_scaling_factor=routed_scaling_factor,
             use_fused_gate=use_fused_gate
         )
 
@@ -495,7 +493,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
         logical_to_physical_map: Optional[torch.Tensor] = None,
         logical_replica_count: Optional[torch.Tensor] = None,
         use_nn_moe: Optional[bool] = False,
-        routed_scaling_factor: Optional[float] = None,
         use_fused_gate: Optional[bool] = False,
     ) -> torch.Tensor:
 
@@ -517,7 +514,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
             expert_load_view=expert_load_view,
             logical_to_physical_map=logical_to_physical_map,
             logical_replica_count=logical_replica_count,
-            routed_scaling_factor=routed_scaling_factor,
             use_fused_gate=use_fused_gate)
 
         if self.rocm_aiter_moe_enabled:
@@ -588,7 +584,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
         logical_to_physical_map: Optional[torch.Tensor] = None,
         logical_replica_count: Optional[torch.Tensor] = None,
         use_nn_moe: Optional[bool] = False,
-        routed_scaling_factor: Optional[float] = None,
         use_fused_gate: Optional[bool] = False,
     ):
         if enable_eplb is not False or expert_load_view is not None or \
@@ -828,7 +823,6 @@ class FusedMoE(CustomOp):
         enable_eplb: bool = False,
         num_redundant_experts: int = 0,
         has_bias: bool = False,
-        routed_scaling_factor: Optional[float] = None,
     ):
         super().__init__()
         if params_dtype is None:
@@ -909,7 +903,6 @@ class FusedMoE(CustomOp):
         self.e_score_correction_bias = e_score_correction_bias
         self.apply_router_weight_on_input = apply_router_weight_on_input
         self.activation = activation
-        self.routed_scaling_factor = routed_scaling_factor
 
         if self.scoring_func != "softmax" and not self.use_grouped_topk:
             raise ValueError("Only softmax scoring function is supported for "
@@ -1519,7 +1512,6 @@ class FusedMoE(CustomOp):
         expert_load_view: Optional[torch.Tensor] = None,
         logical_to_physical_map: Optional[torch.Tensor] = None,
         logical_replica_count: Optional[torch.Tensor] = None,
-        routed_scaling_factor: Optional[float] = None,
         use_fused_gate: Optional[bool] = False
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
@@ -1827,7 +1819,6 @@ class FusedMoE(CustomOp):
             logical_to_physical_map=self.logical_to_physical_map,
             logical_replica_count=self.logical_replica_count,
             use_nn_moe=self.use_nn_moe,
-            routed_scaling_factor=self.routed_scaling_factor,
             use_fused_gate=self.use_fused_gate
         )
 

vllm/model_executor/models/qwen3.py

@@ -38,11 +38,9 @@ from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (QKVParallelLinear,
                                                RowParallelLinear)
 from vllm.model_executor.layers.logits_processor import LogitsProcessor
-from vllm.model_executor.layers.pooler import Pooler, PoolingType
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead
-from vllm.model_executor.pooling_metadata import PoolingMetadata
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors, PoolerOutput