Update app.py

app.py

@@ -252,42 +252,43 @@ def regenerate(state, image_process_mode):
 @spaces.GPU
 def get_interm_outs(state):
     prompt = state.get_prompt()
+    print(prompt)
     images = state.get_images(return_pil=True)
     #prompt, image_args = process_image(prompt, images)
 
-    if images is not None and len(images) > 0:
-        if len(images) > 0:
-            if len(images) != prompt.count(DEFAULT_IMAGE_TOKEN):
-                raise ValueError("Number of images does not match number of <image> tokens in prompt")
+    # if images is not None and len(images) > 0:
+    #     if len(images) > 0:
+    #         if len(images) != prompt.count(DEFAULT_IMAGE_TOKEN):
+    #             raise ValueError("Number of images does not match number of <image> tokens in prompt")
             
-            #images = [load_image_from_base64(image) for image in images]
-            image_sizes = [image.size for image in images]
-            inp_images = process_images(images, image_processor, model.config)
-
-            if type(inp_images) is list:
-                inp_images = [image.to(model.device, dtype=torch.float16) for image in images]
-            else:
-                inp_images = inp_images.to(model.device, dtype=torch.float16)
-        else:
-            inp_images = None
-            image_sizes = None
-        image_args = {"images": inp_images, "image_sizes": image_sizes}
-    else:
-        inp_images = None
-        image_args = {}
-
-    input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0).to(model.device)
-
-    interm_outs = model.get_visual_interpretations(
-                input_ids,
-                **image_args
-        )
+    #         #images = [load_image_from_base64(image) for image in images]
+    #         image_sizes = [image.size for image in images]
+    #         inp_images = process_images(images, image_processor, model.config)
+
+    #         if type(inp_images) is list:
+    #             inp_images = [image.to(model.device, dtype=torch.float16) for image in images]
+    #         else:
+    #             inp_images = inp_images.to(model.device, dtype=torch.float16)
+    #     else:
+    #         inp_images = None
+    #         image_sizes = None
+    #     image_args = {"images": inp_images, "image_sizes": image_sizes}
+    # else:
+    #     inp_images = None
+    #     image_args = {}
+
+    # input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0).to(model.device)
+
+    # interm_outs = model.get_visual_interpretations(
+    #             input_ids,
+    #             **image_args
+    #     )
     
-    depth_outs = get_depth_images(interm_outs, image_sizes[0]) 
-    seg_outs =  get_seg_images(interm_outs, images[0])
-    gen_outs = get_gen_images(interm_outs)
+    # depth_outs = get_depth_images(interm_outs, image_sizes[0]) 
+    # seg_outs =  get_seg_images(interm_outs, images[0])
+    # gen_outs = get_gen_images(interm_outs)
 
-    return depth_outs, seg_outs, gen_outs
+    return images[0], images[0], images[0]
     
 
 @spaces.GPU