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

python/sglang/srt/models/llama_reward.py

@@ -18,7 +18,6 @@ from typing import Iterable, Optional, Tuple
 import torch
 from torch import nn
 from transformers import LlamaConfig
-from vllm.config import CacheConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessorOutput
@@ -33,7 +32,7 @@ class LlamaForSequenceClassification(nn.Module):
         self,
         config: LlamaConfig,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
     ) -> None:
         super().__init__()
         self.config = config
@@ -92,7 +91,7 @@ class LlamaForSequenceClassificationWithNormal_Weights(LlamaForSequenceClassific
         self,
         config: LlamaConfig,
         quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
+        cache_config=None,
     ) -> None:
         super().__init__(config, quant_config, cache_config)
         self.weights = self.Weights(config.hidden_size, self.num_labels)