fix other vlm

Author: sangchengmeng

lightllm/models/gemma3/gemma3_visual.py

@@ -17,7 +17,6 @@
 
 class Gemma3VisionModel:
     def __init__(self, kvargs):
-        self.remote_vit = kvargs.get("remote_vit", False)
         pass
 
     def load_model(self, weight_dir):
@@ -123,10 +122,7 @@ def encode(self, images: List[ImageItem]):
         for i, img in enumerate(images):
             if isinstance(img, ImageItem):
                 uuids.append(img.uuid)
-                if self.remote_vit:
-                    image_data = img._preload_data
-                else:
-                    image_data = read_shm(get_shm_name_data(img.uuid))
+                image_data = read_shm(get_shm_name_data(img.uuid))
                 image_data = Image.open(BytesIO(image_data))
                 t = self.image_processor.preprocess(image_data, return_tensors="pt")["pixel_values"]
                 img_tensors.append(t)

lightllm/models/llava/llava_visual.py

@@ -16,7 +16,6 @@
 
 class LlavaVisionModel:
     def __init__(self, kvargs):
-        self.remote_vit = kvargs.get("remote_vit", False)
         pass
 
     def load_model(self, weight_dir):

lightllm/models/qwen2_5_vl/qwen2_5_visual.py

@@ -171,7 +171,6 @@ def __init__(
         self.window_size = window_size
         self.fullatt_block_indexes = fullatt_block_indexes
         self.out_hidden_size = out_hidden_size
-        self.remote_vit = kvargs.get("remote_vit", False)
 
         self.spatial_merge_unit = self.spatial_merge_size * self.spatial_merge_size
 

lightllm/models/qwen2_vl/qwen2_visual.py

@@ -200,7 +200,6 @@ def __init__(
         self.patch_size = patch_size
         self.spatial_merge_size = spatial_merge_size
         self.temporal_patch_size = temporal_patch_size
-        self.remote_vit = kvargs.get("remote_vit", False)
 
         self.patch_embed = PatchEmbed(
             patch_size=self.patch_size,

lightllm/models/qwen_vl/qwen_visual.py

@@ -345,7 +345,6 @@ def __init__(
         **kwargs,
     ):
         super().__init__()
-        self.remote_vit = kvargs.get("remote_vit", False)
         image_height, image_width = self.image_size = (image_size, image_size)
         patch_height, patch_width = self.patch_size = (patch_size, patch_size)
         self.grid_size = (image_height // patch_height, image_width // patch_width)
@@ -424,10 +423,7 @@ def encode(self, image_uuids: List):
         for i, item in enumerate(image_uuids):
             if isinstance(item, int):
                 uuids.append(item)
-                if self.remote_vit:
-                    image_data = item._preload_data
-                else:
-                    image_data = read_shm(get_shm_name_data(item.uuid))
+                image_data = read_shm(get_shm_name_data(item.uuid))
                 image_data = Image.open(BytesIO(image_data)).convert("RGB")
                 t = self.image_transform(image_data)
                 img_tensors.append(t)

lightllm/models/tarsier2/tarsier2_visual.py

@@ -166,7 +166,6 @@ def __init__(
         **kwargs,
     ):
         super().__init__()
-        self.remote_vit = kvargs.get("remote_vit", False)
         self.vision_tower = Qwen2VisionTransformerPretrainedModel(**vision_config)
 
         if projection_head == "Pixel_Shuffle":

lightllm/models/vit/model.py

@@ -48,7 +48,6 @@ def __init__(self, kvargs):
         self.quant_cfg_path = kvargs.get("quant_cfg", None)
         self.load_image_func = get_load_image_func(self.weight_dir_)
         self.max_batch_size = kvargs.get("max_batch_size", 1)
-        self.remote_vit = kvargs.get("remote_vit", False)
 
         self._init_datatype()
         self._init_config()