Remove unused kwargs from model definitions (#13555)

vllm/model_executor/models/chatglm.py

@@ -2,13 +2,13 @@
 # Adapted from
 # https://github.com/THUDM/ChatGLM2-6B
 """Inference-only ChatGLM model compatible with THUDM weights."""
-from typing import Iterable, List, Optional, Set, Tuple, Union
+from typing import Iterable, Optional, Set, Tuple, Union
 
 import torch
 from torch import nn
 from torch.nn import LayerNorm
 
-from vllm.attention import Attention, AttentionMetadata
+from vllm.attention import Attention
 from vllm.config import CacheConfig, VllmConfig
 from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import SiluAndMul
@@ -108,19 +108,11 @@ class GLMAttention(nn.Module):
         self,
         hidden_states: torch.Tensor,
         position_ids: torch.Tensor,
-        kv_cache: torch.Tensor,
-        attn_metadata: AttentionMetadata,
     ) -> torch.Tensor:
         qkv, _ = self.query_key_value(hidden_states)
         q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
         q, k = self.rotary_emb(position_ids, q, k)
-        context_layer = self.attn(
-            q,
-            k,
-            v,
-            kv_cache,
-            attn_metadata,
-        )
+        context_layer = self.attn(q, k, v)
         attn_output, _ = self.dense(context_layer)
         return attn_output
 
@@ -215,8 +207,6 @@ class GLMBlock(nn.Module):
         self,
         hidden_states: torch.Tensor,
         position_ids: torch.Tensor,
-        kv_cache: torch.Tensor,
-        attn_metadata: AttentionMetadata,
     ) -> torch.Tensor:
         # hidden_states: [num_tokens, h]
         # Layer norm at the beginning of the transformer layer.
@@ -225,8 +215,6 @@ class GLMBlock(nn.Module):
         attention_output = self.self_attention(
             hidden_states=layernorm_output,
             position_ids=position_ids,
-            kv_cache=kv_cache,
-            attn_metadata=attn_metadata,
         )
 
         # Residual connection.
@@ -289,17 +277,10 @@ class GLMTransformer(nn.Module):
         self,
         hidden_states: torch.Tensor,
         position_ids: torch.Tensor,
-        kv_caches: List[torch.Tensor],
-        attn_metadata: AttentionMetadata,
     ) -> Union[torch.Tensor, IntermediateTensors]:
-        for i in range(self.start_layer, self.end_layer):
-            layer = self.layers[i]
-            hidden_states = layer(
-                hidden_states=hidden_states,
-                position_ids=position_ids,
-                kv_cache=kv_caches[i - self.start_layer],
-                attn_metadata=attn_metadata,
-            )
+        for layer in self.layers[self.start_layer:self.end_layer]:
+            hidden_states = layer(hidden_states=hidden_states,
+                                  position_ids=position_ids)
 
         if not get_pp_group().is_last_rank:
             return IntermediateTensors({"hidden_states": hidden_states})
@@ -350,8 +331,6 @@ class ChatGLMModel(nn.Module):
         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,
@@ -369,8 +348,6 @@ class ChatGLMModel(nn.Module):
         hidden_states = self.encoder(
             hidden_states=hidden_states,
             position_ids=positions,
-            kv_caches=kv_caches,
-            attn_metadata=attn_metadata,
         )
 
         return hidden_states
@@ -494,12 +471,9 @@ class ChatGLMForCausalLM(ChatGLMBaseModel, SupportsLoRA, SupportsPP):
         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,
     ) -> Union[torch.Tensor, IntermediateTensors]:
-        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