Update app_dialogue.py

app_dialogue.py

@@ -75,23 +75,33 @@ logger = logging.getLogger()
 
 # Monkey patch adapted from gradio.components.image.Image - mostly to make the `save` step optional in `pil_to_temp_file`
 def hash_bytes(bytes: bytes):
+    print("********* hash_bytes ********")
+    print(f"param: bytes - {bytes}")
     sha1 = hashlib.sha1()
     sha1.update(bytes)
+    print(f"Returns - sha1.hexdigest() - {sha1.hexdigest()}")
     return sha1.hexdigest()
 
 
 def pil_to_temp_file(img: PIL.Image.Image, dir: str = DEFAULT_TEMP_DIR, format: str = "png") -> str:
-    """Save a PIL image into a temp file"""
+    """
+    Save a PIL image into a temp file
+    """
+    print("************ pil_to_temp_file ***********")
+    print(f"img - {img}, dir - {dir}, format - {format}")
     bytes_data = processing_utils.encode_pil_to_bytes(img, format)
     temp_dir = Path(dir) / hash_bytes(bytes_data)
     temp_dir.mkdir(exist_ok=True, parents=True)
     filename = str(temp_dir / f"image.{format}")
     if not os.path.exists(filename):
         img.save(filename, pnginfo=processing_utils.get_pil_metadata(img))
+    print(f"filename - {filename}")
     return filename
 
 
 def add_file(file):
+    print("****** add_file *******")
+    print(f"returns - file name - {file.name}")
     return file.name
 
 
@@ -103,6 +113,8 @@ def split_str_on_im_markdown(string: str) -> List[str]:
     - `User:![](https://favurl.com/chicken_on_money.png)Describe this image.` would become `["User:", "https://favurl.com/chicken_on_money.png", "Describe this image."]`
     - `User:![](/file=/my_temp/chicken_on_money.png)Describe this image.` would become `["User:", "/my_temp/chicken_on_money.png", "Describe this image."]`
     """
+    print("************ split_str_on_im_markdown ***********")
+    print(f"string - {string}")
     IMAGES_PATTERN = re.compile(r"!\[[^\]]*\]\((.*?)\s*(\"(?:.*[^\"])\")?\s*\)")
     parts = []
     cursor = 0
@@ -117,6 +129,8 @@ def split_str_on_im_markdown(string: str) -> List[str]:
         cursor = pattern.end()
     if cursor != len(string):
         parts.append(string[cursor:])
+
+    print(f"returns - parts - {parts}")
     return parts
 
 
@@ -124,6 +138,9 @@ def is_image(string: str) -> bool:
     """
     There are two ways for images: local image path or url.
     """
+    print("************ is_image ***********")
+    print(f"string - {string}")
+    print(f"returns - a bool - {is_url(string) or string.startswith(DEFAULT_TEMP_DIR)}")
     return is_url(string) or string.startswith(DEFAULT_TEMP_DIR)
 
 
@@ -132,9 +149,12 @@ def is_url(string: str) -> bool:
     Checks if the passed string contains a valid url and nothing else. e.g. if space is included it's immediately
     invalidated the url
     """
+    print("************ is_url ***********")
+    print(f"string - {string}")
     if " " in string:
         return False
     result = urlparse(string)
+    print(f"returns - a bool - {all([result.scheme, result.netloc])}")
     return all([result.scheme, result.netloc])
 
 
@@ -161,6 +181,8 @@ def isolate_images_urls(prompt_list: List) -> List:
     ]
     ```
     """
+    print("************ isolate_images_urls ***********")
+    print(f"params: prompt_list - {prompt_list}")
     linearized_list = []
     for prompt in prompt_list:
         # Prompt can be either a string, or a PIL image
@@ -183,11 +205,15 @@ def isolate_images_urls(prompt_list: List) -> List:
                 f"Unrecognized type for `prompt`. Got {type(type(prompt))}. Was expecting something in [`str`,"
                 " `PIL.Image.Image`]"
             )
+    print(f"linearized_list - {linearized_list}")
     return linearized_list
 
 
 def fetch_images(url_list: str) -> PIL.Image.Image:
     """Fetching images"""
+    print("************ fetch_images ***********")
+    print(f"params: url_list - {url_list}")
+    print(f"returns - PROCESSOR.image_processor.fetch_images(url_list) - {PROCESSOR.image_processor.fetch_images(url_list)}")
     return PROCESSOR.image_processor.fetch_images(url_list)
 
 
@@ -196,6 +222,8 @@ def handle_manual_images_in_user_prompt(user_prompt: str) -> List[str]:
     Handle the case of textually manually inputted images (i.e. the `<fake_token_around_image><image:IMG_URL><fake_token_around_image>`) in the user prompt
     by fetching them, saving them locally and replacing the whole sub-sequence the image local path.
     """
+    print("************ handle_manual_images_in_user_prompt ***********")
+    print(f"params: user_prompt - {user_prompt}")
     if "<fake_token_around_image>" in user_prompt:
         splitted_user_prompt = isolate_images_urls([user_prompt])
         resulting_user_prompt = []
@@ -206,13 +234,18 @@ def handle_manual_images_in_user_prompt(user_prompt: str) -> List[str]:
                 resulting_user_prompt.append(tmp_file)
             else:
                 resulting_user_prompt.append(u_p)
+        print(f"returns - resulting_user_prompt - {resulting_user_prompt}")
         return resulting_user_prompt
     else:
+        print(f"returns - [user_prompt] - {[user_prompt]}")
         return [user_prompt]
 
 
 def gradio_link(img_path: str) -> str:
+    print("************ gradio_link ***********")
+    print(f"params: img_path - {img_path}")
     url = f"{GRADIO_LINK}/file={img_path}"
+    print(f"returns - url - {url}")
     return url
 
 
@@ -221,6 +254,8 @@ def prompt_list_to_markdown(prompt_list: List[str]) -> str:
     Convert a user prompt in the list format (i.e. elements are either a PIL image or a string) into
     the markdown format that is used for the chatbot history and rendering.
     """
+    print("************ prompt_list_to_markdown ***********")
+    print(f"params: prompt_list - {prompt_list}")
     resulting_string = ""
     for elem in prompt_list:
         if is_image(elem):
@@ -230,6 +265,7 @@ def prompt_list_to_markdown(prompt_list: List[str]) -> str:
                 resulting_string += f"![](/file={elem})"
         else:
             resulting_string += elem
+    print(f"returns - resulting_string - {resulting_string}")
     return resulting_string
 
 
@@ -238,6 +274,8 @@ def prompt_list_to_tgi_input(prompt_list: List[str]) -> str:
     TGI expects a string that contains both text and images in the image markdown format (i.e. the `![]()` ).
     The images links are parsed on TGI side
     """
+    print("************ prompt_list_to_tgi_input ***********")
+    print(f"params: prompt_list - {prompt_list}")
     result_string_input = ""
     for elem in prompt_list:
         if is_image(elem):
@@ -247,13 +285,17 @@ def prompt_list_to_tgi_input(prompt_list: List[str]) -> str:
                 result_string_input += f"![]({gradio_link(img_path=elem)})"
         else:
             result_string_input += elem
+    print(f"returns - result_string_input - {result_string_input}")
     return result_string_input
 
 
 def remove_spaces_around_token(text: str) -> str:
+    print("************ remove_spaces_around_token ***********")
+    print(f"params: text - {text}")
     pattern = r"\s*(<fake_token_around_image>)\s*"
     replacement = r"\1"
     result = re.sub(pattern, replacement, text)
+    print(f"returns - result - {result}")
     return result
 
 
@@ -265,6 +307,10 @@ 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("************ format_user_prompt_with_im_history_and_system_conditioning ***********")
+    print(f"params: current_user_prompt_str - {current_user_prompt_str}")
+    print(f"params: current_image - {current_image}")
+    print(f"params: history - {history}")
     resulting_list = copy.deepcopy(SYSTEM_PROMPT)
 
     # Format history
@@ -300,6 +346,7 @@ def format_user_prompt_with_im_history_and_system_conditioning(
         resulting_list.extend(["\nUser:", current_image, f"{current_user_prompt_str}<end_of_utterance>\nAssistant:"])
         current_user_prompt_list = [current_user_prompt_str]
 
+    print(f"returns - resulting_list - {resulting_list}, current_user_prompt_list - {current_user_prompt_list}")
     return resulting_list, current_user_prompt_list
 
 
@@ -472,6 +519,16 @@ with gr.Blocks(title="IDEFICS Playground", theme=gr.themes.Base()) as demo:
         repetition_penalty,
         top_p,
     ):
+    print("************ model_inference ***********")
+    print(f"params: model_selector - {model_selector}")
+    print(f"params: user_prompt_str - {user_prompt_str}")
+    print(f"params: chat_history - {chat_history}")
+    print(f"params: image - {image}")
+    print(f"params: decoding_strategy - {decoding_strategy}")
+    print(f"params: temperature - {temperature}")
+    print(f"params: max_new_tokens - {max_new_tokens}")
+    print(f"params: repetition_penalty - {repetition_penalty}")
+    print(f"params: top_p - {top_p}")
         if user_prompt_str.strip() == "" and image is None:
             return "", None, chat_history
 
@@ -542,6 +599,8 @@ with gr.Blocks(title="IDEFICS Playground", theme=gr.themes.Base()) as demo:
             last_turn = chat_history.pop(-1)
             last_turn[-1] += acc_text
             chat_history.append(last_turn)
+            print(f"yields - chat_history -{chat_history} ")
+            print(f"acc_text -{acc_text} ")
             yield "", None, chat_history
             acc_text = ""
 
@@ -557,6 +616,7 @@ with gr.Blocks(title="IDEFICS Playground", theme=gr.themes.Base()) as demo:
             repetition_penalty=None,
             top_p=0.95,
         )
+        print("************ process_example ***********")
         return clear_msg, image_value, chat
 
     textbox.submit(
@@ -595,6 +655,8 @@ with gr.Blocks(title="IDEFICS Playground", theme=gr.themes.Base()) as demo:
     )
 
     def remove_last_turn(chat_history):
+        print("************ remove_last_turn ***********")
+        print(f"params: chat_history - {chat_history}")
         if len(chat_history) == 0:
             return gr.Update(), gr.Update()
         last_interaction = chat_history[-1]