Update app_dialogue.py

app_dialogue.py

@@ -115,7 +115,26 @@ def convert_to_rgb(filepath_or_pilimg):
 
         return temp_file_path  # Return the path to the saved image
 
+def pil_to_markdown_im(image):
+    """
+    Convert a PIL image into markdown filled with the base64 string representation.
+    """
+    print(f"***** pil_to_markdown_im ******")
+    print(f"params: image is - {image}")
+    #if isinstance(image, PIL.Image.Image):
+    #img_b64_str = pil_to_base64(image)
+    #img_str = f'<img src="data:image/png;base64,{img_b64_str}" />'
+    #if path_or_url.startswith(("http://", "https://")):
+    #response = requests.get(image)
+    #image = Image.open(BytesIO(response.content))
+    # Generate a unique filename using UUID
+    filename = f"{uuid.uuid4()}.jpg"
+    local_path = f"{filename}" 
+    image.save(local_path)
+    img_str =  f"![](/file={local_path})"
+    return img_str
 
+    
 def base64_to_pil(encoded_image):
     decoded_image = base64.b64decode(encoded_image)
     pil_image = Image.open(BytesIO(decoded_image))
@@ -322,37 +341,60 @@ def format_user_prompt_with_im_history_and_system_conditioning(
     Produces the resulting list that needs to go inside the processor.
     It handles the potential image box input, the history and the system conditionning.
     """
+    print(f"*********format_user_prompt_with_im_history_and_system_conditioning*********")
+    print(f"format_user_prompt_with_im_history_and_system_conditioning -- param current_user_prompt_str is - {current_user_prompt_str} ")
+    print(f"format_user_prompt_with_im_history_and_system_conditioning -- param current_image is - {current_image} ")
+    print(f"format_user_prompt_with_im_history_and_system_conditioning -- param history is - {history} ")
+    
     resulting_list = copy.deepcopy(SYSTEM_PROMPT)
 
     # Format history
     for turn in history:
+        print(f"inside for loop, turn is - {turn}")
         user_utterance, assistant_utterance = turn
+        print("calling split_str_on_im_markdown from inside for loop inside format_user_prompt_with_im_history_and_system_conditioning")
         splitted_user_utterance = split_str_on_im_markdown(user_utterance)
+        print(f"splitted_user_utterance from split_str_on_im_markdown is - {splitted_user_utterance} ")
         splitted_user_utterance = [
             im_markdown_to_pil(s) if s.startswith('<img src="data:image/png;base64,') else s
             for s in splitted_user_utterance
             if s != ""
         ]
+        print(f"splitted_user_utterance after im_markdown_to_pil() is - {splitted_user_utterance} ")
+        
         if isinstance(splitted_user_utterance[0], str):
             resulting_list.append("\nUser: ")
         else:
             resulting_list.append("\nUser:")
+        print(f"resulting_list after if..else block is - {resulting_list}")
         resulting_list.extend(splitted_user_utterance)
+        print(f"resulting_list after extend is - {resulting_list}")
         resulting_list.append(f"<end_of_utterance>\nAssistant: {assistant_utterance}")
+        print(f"resulting_list after append is - {resulting_list}")
+        
 
     # Format current input
     current_user_prompt_str = remove_spaces_around_token(current_user_prompt_str)
+    print(f"current_user_prompt_str is - {current_user_prompt_str}")
+    
     if current_image is None:
+        print("inside IF : current_image is NONE")
         if "<img src=data:image/png;base64" in current_user_prompt_str:
             raise ValueError("The UI does not support inputing via the text box an image in base64.")
         current_user_prompt_list = handle_manual_images_in_user_prompt(current_user_prompt_str)
+        print(f"current_user_prompt_list (or [user_prompt]/resulting_user_prompt((most likely this one)) from handle_manual_images_in_user_prompt ) is - {current_user_prompt_list}")
         resulting_list.append("\nUser: ")
+        print(f"resulting_list with append user - {resulting_list}")
         resulting_list.extend(current_user_prompt_list)
+        print(f"resulting_list after extend with current_user_prompt_list is - {resulting_list}")
         resulting_list.append("<end_of_utterance>\nAssistant:")
+        print(f"resulting_list after append with end_of_utteranceAssistant is - {resulting_list}")
         return resulting_list, current_user_prompt_list
     else:
+        print("inside ELSE : current_image is not NONE")
         # Choosing to put the image first when the image is inputted through the UI, but this is an arbiratrary choice.
-        resulting_list.extend(["\nUser:", current_image, f"{current_user_prompt_str}<end_of_utterance>\nAssistant:"])
+        resulting_list.extend(["\nUser:", Image.open(current_image), f"{current_user_prompt_str}<end_of_utterance>\nAssistant:"]) #current_image
+        print(f"final resulting_list passed on to calling function is - {resulting_list}")
         return resulting_list, [current_user_prompt_str]
 
 
@@ -535,7 +577,7 @@ with gr.Blocks(title="IDEFICS-Chat", theme=gr.themes.Base()) as demo:
                 )
             processor, tokenizer, model = load_processor_tokenizer_model(model_selector.value)
 
-            imagebox = gr.Image(type="pil", label="Image input")
+            imagebox = gr.Image(type="filepath", label="Image input")
 
             with gr.Accordion("Advanced parameters", open=False, visible=True) as parameter_row:
                 max_new_tokens = gr.Slider(