Update vllm to v0.4.3 (#511)

Co-authored-by: Qubitium <417764+Qubitium@users.noreply.github.com>
Co-authored-by: ZX <zx@lbx.dev>

python/sglang/srt/models/llavavid.py

@@ -7,6 +7,7 @@ import torch
 from torch import nn
 from transformers import CLIPVisionModel, LlavaConfig
 from transformers.models.llava.modeling_llava import LlavaMultiModalProjector
+from vllm.config import CacheConfig
 from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
@@ -25,6 +26,7 @@ class LlavaVidForCausalLM(nn.Module):
         self,
         config: LlavaConfig,
         quant_config: Optional[QuantizationConfig] = None,
+        cache_config: Optional[CacheConfig] = None,
     ) -> None:
         super().__init__()
         self.config = config