[1/N] Remove `CacheConfig` import in all model files (#1658)

python/sglang/srt/models/mixtral_quant.py

@@ -23,7 +23,6 @@ import torch
 import torch.nn.functional as F
 from torch import nn
 from transformers import MixtralConfig
-from vllm.config import CacheConfig
 from vllm.distributed import (
     get_tensor_model_parallel_rank,
     get_tensor_model_parallel_world_size,
@@ -325,7 +324,7 @@ class QuantMixtralForCausalLM(nn.Module):
         self,
         config: MixtralConfig,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
     ) -> None:
         super().__init__()
         self.config = config

python/sglang/srt/models/olmoe.py

@@ -23,7 +23,6 @@ import torch
 import torch.nn.functional as F
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.config import CacheConfig
 from vllm.distributed import (
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
@@ -298,7 +297,7 @@ class OlmoeForCausalLM(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()

python/sglang/srt/models/qwen.py

@@ -20,7 +20,6 @@ from typing import Any, Dict, Iterable, Optional, Tuple
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
@@ -243,7 +242,7 @@ class QWenLMHeadModel(nn.Module):
         self,
         config: PretrainedConfig,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
     ):
         super().__init__()
         self.config = config

python/sglang/srt/models/qwen2.py

@@ -20,7 +20,6 @@ from typing import Any, Dict, Iterable, Optional, Tuple
 
 import torch
 from torch import nn
-from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
@@ -268,7 +267,7 @@ class Qwen2ForCausalLM(nn.Module):
         self,
         config: Qwen2Config,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
     ) -> None:
         super().__init__()
         self.config = config

python/sglang/srt/models/qwen2_moe.py

@@ -23,7 +23,6 @@ import torch
 import torch.nn.functional as F
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.config import CacheConfig
 from vllm.distributed import (
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
@@ -160,7 +159,7 @@ class Qwen2MoeAttention(nn.Module):
         rope_theta: float = 10000,
         rope_scaling: Optional[Dict[str, Any]] = None,
         max_position_embeddings: int = 8192,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
@@ -236,7 +235,7 @@ class Qwen2MoeDecoderLayer(nn.Module):
         self,
         config: PretrainedConfig,
         layer_id: int,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
@@ -306,7 +305,7 @@ class Qwen2MoeModel(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
@@ -355,7 +354,7 @@ class Qwen2MoeForCausalLM(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()

python/sglang/srt/models/stablelm.py

@@ -22,7 +22,6 @@ from typing import Iterable, Optional, Tuple
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
@@ -241,7 +240,7 @@ class StableLmForCausalLM(nn.Module):
         self,
         config: PretrainedConfig,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
     ) -> None:
         super().__init__()
         self.config = config

python/sglang/srt/models/torch_native_llama.py

@@ -24,7 +24,6 @@ import torch
 from torch import nn
 from torch.nn.parameter import Parameter
 from transformers import LlamaConfig
-from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
@@ -380,7 +379,7 @@ class TorchNativeLlamaForCausalLM(nn.Module):
         self,
         config: LlamaConfig,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
     ) -> None:
         super().__init__()
         self.config = config

python/sglang/srt/models/xverse.py

@@ -22,7 +22,6 @@ from typing import Any, Dict, Iterable, Optional, Tuple
 import torch
 from torch import nn
 from transformers import LlamaConfig
-from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.layernorm import RMSNorm
@@ -297,7 +296,7 @@ class XverseForCausalLM(nn.Module):
         self,
         config: LlamaConfig,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         efficient_weight_load=False,
     ) -> None:
         super().__init__()

python/sglang/srt/models/xverse_moe.py

@@ -19,7 +19,6 @@ from typing import Any, Dict, Iterable, Optional, Tuple
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.config import CacheConfig
 from vllm.distributed import (
     get_tensor_model_parallel_rank,
     get_tensor_model_parallel_world_size,
@@ -183,7 +182,7 @@ class XverseAttention(nn.Module):
         rope_theta: float = 10000,
         rope_scaling: Optional[Dict[str, Any]] = None,
         max_position_embeddings: int = 8192,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
@@ -260,7 +259,7 @@ class XverseDecoderLayer(nn.Module):
         self,
         config: PretrainedConfig,
         layer_id: int,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
@@ -328,7 +327,7 @@ class XverseModel(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
@@ -371,7 +370,7 @@ class XverseMoeForCausalLM(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()

python/sglang/srt/models/yivl.py

@@ -20,7 +20,6 @@ from typing import Iterable, Optional, Tuple
 import torch
 import torch.nn as nn
 from transformers import CLIPVisionModel, LlavaConfig
-from vllm.config import CacheConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.quantization.base_config import QuantizationConfig
@@ -32,7 +31,7 @@ class YiVLForCausalLM(LlavaLlamaForCausalLM):
         self,
         config: LlavaConfig,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
     ) -> None:
         super().__init__(config, quant_config, cache_config)