Load vision tower weights of llava if they are present in the checkpo…

…ints.

python/sglang/srt/models/llava.py

@@ -263,13 +263,14 @@ def load_weights(
  projector_weights = {
  "model.mm_projector.0": "multi_modal_projector.linear_1",
  "model.mm_projector.2": "multi_modal_projector.linear_2",
+ "model.vision_tower.vision_tower": "vision_tower", # Update the vision tower weights if we find them in the checkpoint (it may be finetuned).
  }
  params_dict = dict(self.named_parameters())
  for name, loaded_weight in hf_model_weights_iterator(
  model_name_or_path, cache_dir, load_format, revision
  ):
  # FIXME: why projector weights read two times?
- if "projector" in name:
+ if "projector" in name or "vision_tower" in name:
  for weight_name, param_name in projector_weights.items():
  if weight_name in name:
  name = name.replace(weight_name, param_name)