Update modeling_GOT.py

modeling_GOT.py

@@ -484,7 +484,7 @@ class GOTQwenForCausalLM(Qwen2ForCausalLM):
         setattr(torch.nn.Linear, "reset_parameters", lambda self: None)
         setattr(torch.nn.LayerNorm, "reset_parameters", lambda self: None)
 
-    def chat(self, tokenizer, image_file, ocr_type, ocr_box='', ocr_color='', render=False, save_render_file=None, print_prompt=False):
+    def chat(self, tokenizer, image_file, ocr_type, ocr_box='', ocr_color='', render=False, save_render_file=None, print_prompt=False, gradio_input=False):
 
         self.disable_torch_init()
 
@@ -495,7 +495,10 @@ class GOTQwenForCausalLM(Qwen2ForCausalLM):
 
         image_token_len = 256
 
-        image = self.load_image(image_file)
+        if gradio_input:
+            image = image_file.copy()
+        else:
+            image = self.load_image(image_file)
 
         w, h = image.size
         
@@ -713,7 +716,7 @@ class GOTQwenForCausalLM(Qwen2ForCausalLM):
         return processed_images
 
 
-    def chat_crop(self, tokenizer, image_file, render=False, save_render_file=None, print_prompt=False):
+    def chat_crop(self, tokenizer, image_file, render=False, save_render_file=None, print_prompt=False, gradio_input=False):
         # Model
         self.disable_torch_init()
         multi_page=False
@@ -749,7 +752,10 @@ class GOTQwenForCausalLM(Qwen2ForCausalLM):
 
         else:
             qs = 'OCR with format upon the patch reference: '
-            img = self.load_image(image_file)
+            if gradio_input:
+                img = image_file.copy()
+            else:
+                img = self.load_image(image_file)
             sub_images = self.dynamic_preprocess(img)
             ll = len(sub_images)