Remove unused kwargs from model definitions (#13555)

vllm/model_executor/models/qwen_vl.py

@@ -22,7 +22,6 @@ from transformers import (BatchFeature, PretrainedConfig, PreTrainedTokenizer,
 from transformers.image_utils import ImageInput
 from transformers.tokenization_utils_base import TextInput
 
-from vllm.attention import AttentionMetadata
 from vllm.config import VllmConfig
 from vllm.model_executor.layers.activation import get_act_fn
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
@@ -766,8 +765,6 @@ class QwenVLForConditionalGeneration(QWenBaseModel, SupportsPP, SupportsLoRA,
         self,
         input_ids: torch.Tensor,
         positions: torch.Tensor,
-        kv_caches: List[torch.Tensor],
-        attn_metadata: AttentionMetadata,
         intermediate_tensors: Optional[IntermediateTensors] = None,
         inputs_embeds: Optional[torch.Tensor] = None,
         **kwargs: object,
@@ -783,7 +780,6 @@ class QwenVLForConditionalGeneration(QWenBaseModel, SupportsPP, SupportsLoRA,
                                                       vision_embeddings)
             input_ids = None
 
-        hidden_states = self.transformer(input_ids, positions, kv_caches,
-                                         attn_metadata, intermediate_tensors,
-                                         inputs_embeds)
+        hidden_states = self.transformer(input_ids, positions,
+                                         intermediate_tensors, inputs_embeds)
         return hidden_states