Update app.py

app.py

@@ -1,236 +1,39 @@
-import copy
-import PIL
-import re
-import os
-import gradio as gr
-from gradio import processing_utils
-from gradio_client.client import DEFAULT_TEMP_DIR
+import torch
+from transformers import IdeficsForVisionText2Text, AutoProcessor
 
-from typing import List, Optional, Tuple
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
-from huggingface_hub import InferenceClient
+checkpoint = "HuggingFaceM4/idefics-9b-instruct"
+model = IdeficsForVisionText2Text.from_pretrained(checkpoint, torch_dtype=torch.bfloat16).to(device)
+processor = AutoProcessor.from_pretrained(checkpoint)
 
-GENERATE_PROMPT=""
+# We feed to the model an arbitrary sequence of text strings and images. Images can be either URLs or PIL Images.
+prompts = [
+    [
+        "User: What is in this image?",
+        "https://upload.wikimedia.org/wikipedia/commons/8/86/Id%C3%A9fix.JPG",
+        "<end_of_utterance>",
 
-client = InferenceClient("HuggingFaceM4/idefics-9b-instruct")
+        "\nAssistant: This picture depicts Idefix, the dog of Obelix in Asterix and Obelix. Idefix is running on the ground.<end_of_utterance>",
 
-chat_models=[
-    "HuggingFaceM4/idefics-9b-instruct",
-    "HuggingFaceM4/idefics-80b-instruct",
-]
-EOS_STRINGS = ["<end_of_utterance>", "\nUser:"]
-SYSTEM_PROMPT = [
-    """The following is a conversation between a highly knowledgeable and intelligent visual AI assistant, called Assistant, and a human user, called User. In the following interactions, User and Assistant will converse in natural language, and Assistant will do its best to answer User’s questions. Assistant has the ability to perceive images and reason about the content of visual inputs. Assistant was built to be respectful, polite and inclusive. It knows a lot, and always tells the truth. When prompted with an image, it does not make up facts.
-The conversation begins:""",
-    """\nUser:""",
-    "https://i1.pickpik.com/photos/515/906/355/kittens-cats-pet-cute-preview.jpg",
-    "Describe this image.<end_of_utterance>",
-    """\nAssistant: Five kittens are standing together in the center of the photograph. From the left to right, there is one orange kitten, two kittens white and black stripes, and two brown kittens with an orange head. They are in bright green grass and it looks like they are moving forward.<end_of_utterance>""",
-    "\nUser:How about this image?",
-    "https://cdn.pixabay.com/photo/2017/09/25/13/12/puppy-2785074_1280.jpg",
-    "Can you describe it too?<end_of_utterance>",
-    """\nAssistant: A dog is lying on the floor, looking at the camera. It is looking directly at you. The dog has a white body and brown patches on its face and ears. Its eyes are dark. Its nose is black, and it has long, floppy ears, white paws, long fur, big eyes.<end_of_utterance>""",
-    "\nUser: What kind of breed is it?<end_of_utterance>",
-    """\nAssistant: It looks like a Cavalier King Charles Spaniel. These breeds are known for their gentle, affectionate nature. Their long ears and big, brown eyes give them a sweet, expressive face. Cavaliers are small dogs weighing between 13 and 20 pounds. They have a silky, medium-length coat that comes in various colors, including black and tan, tricolor, red and white, or black.<end_of_utterance>""",
-    "\nUser: What can you tell me about this breed of dogs?<end_of_utterance>",
-    """\nAssistant: One specific characteristic of the Cavalier King Charles Spaniel is its friendly and affectionate nature. This breed is known for being extremely sociable and forming strong bonds with their owners. They are often described as "velcro dogs" because they love to be close to their human companions, whether it's sitting on their laps, cuddling on the couch, or simply following them around the house.<end_of_utterance>""",
-    "\nUser: ghjkhjabnufs<end_of_utterance>",
-    """\nAssistant: That doesn’t seem to be a word. Could you ask me another way?<end_of_utterance>""",
-    "\nUser: Do you like Cavalier King Charles Spaniel?<end_of_utterance>",
-    """\nAssistant: I do not have personal opinions as I’m just a computer program. However, cavaliers are known for being excellent family pets due to their gentle and patient demeanor, making them great with children and other pets. Their affectionate nature also makes them ideal therapy dogs, as they can provide comfort and emotional support to those in need.<end_of_utterance>""",
-    "\nUser: How many dogs do you see in this image?",
-    "https://i.dailymail.co.uk/i/pix/2011/07/01/article-2010308-0CD22A8300000578-496_634x414.jpg",
-    "<end_of_utterance>",
-    """\nAssistant: There is no dogs in this image. The picture shows a tennis player jumping to volley the ball.<end_of_utterance>""",
-]
-def run_gpt(in_prompt,history,model_drop=None):
-    #client = InferenceClient(chat_models.models[int(model_drop)])
-    print(f'history :: {history}')
-    prompt=format_prompt(in_prompt,history)
-    seed = random.randint(1,1111111111111111)
-    print (seed)
-    generate_kwargs = dict(
-        temperature=1.0,
-        max_new_tokens=1048,
-        top_p=0.99,
-        repetition_penalty=1.0,
-        do_sample=True,
-        seed=seed,
-    )
-    content = GENERATE_PROMPT + prompt
-    print(content)
-    stream = client.text_generation(content, **generate_kwargs, stream=True, details=True, return_full_text=False)
-    resp = ""
-    for response in stream:
-        resp += response.token.text
-    return resp
-
-
-
-def hash_bytes(bytes: bytes):
-    sha1 = hashlib.sha1()
-    sha1.update(bytes)
-    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"""
-    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))
-    return filename
-def add_file(file):
-    return file.name, gr.update(label='🖼️ Uploaded!')
-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.
-    """
-    resulting_string = ""
-    for elem in prompt_list:
-        if is_image(elem):
-            if is_url(elem):
-                resulting_string += f"![]({elem})"
-            else:
-                resulting_string += f"![](/file={elem})"
-        else:
-            resulting_string += elem
-    return resulting_string
-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
-    """
-    result_string_input = ""
-    for elem in prompt_list:
-        if is_image(elem):
-            if is_url(elem):
-                result_string_input += f"![]({elem})"
-            else:
-                result_string_input += f"![]({gradio_link(img_path=elem)})"
-        else:
-            result_string_input += elem
-    return result_string_input
-def remove_spaces_around_token(text: str) -> str:
-    pattern = r"\s*(<fake_token_around_image>)\s*"
-    replacement = r"\1"
-    result = re.sub(pattern, replacement, text)
-    return result
-def format_user_prompt_with_im_history_and_system_conditioning(
-    current_user_prompt_str: str, current_image: Optional[str], history: List[Tuple[str, str]]
-) -> Tuple[List[str], List[str]]:
-    """
-    Produces the resulting list that needs to go inside the processor.
-    It handles the potential image box input, the history and the system conditionning.
-    """
-    resulting_list = copy.deepcopy(SYSTEM_PROMPT)
-
-    # Format history
-    for turn in history:
-        user_utterance, assistant_utterance = turn
-        splitted_user_utterance = split_str_on_im_markdown(user_utterance)
-
-        optional_space = ""
-        if not is_image(splitted_user_utterance[0]):
-            optional_space = " "
-        resulting_list.append(f"\nUser:{optional_space}")
-        resulting_list.extend(splitted_user_utterance)
-        resulting_list.append(f"<end_of_utterance>\nAssistant: {assistant_utterance}")
-
-    # Format current input
-    current_user_prompt_str = remove_spaces_around_token(current_user_prompt_str)
-    if current_image is None:
-        if "![](" in current_user_prompt_str:
-            current_user_prompt_list = split_str_on_im_markdown(current_user_prompt_str)
-        else:
-            current_user_prompt_list = handle_manual_images_in_user_prompt(current_user_prompt_str)
+        "\nUser:",
+        "https://static.wikia.nocookie.net/asterix/images/2/25/R22b.gif/revision/latest?cb=20110815073052",
+        "And who is that?<end_of_utterance>",
 
-        optional_space = ""
-        if not is_image(current_user_prompt_list[0]):
-            # Check if the first element is an image (and more precisely a path to an image)
-            optional_space = " "
-        resulting_list.append(f"\nUser:{optional_space}")
-        resulting_list.extend(current_user_prompt_list)
-        resulting_list.append("<end_of_utterance>\nAssistant:")
-    else:
-        # 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:"])
-        current_user_prompt_list = [current_user_prompt_str]
-
-    return resulting_list, current_user_prompt_list
-
-def process_example(message, image):
-    """
-    Same as `model_inference` but in greedy mode and with the 80b-instruct.
-    Specifically for pre-computing the default examples.
-    """
-    #model_selector="HuggingFaceM4/idefics-80b-instruct"
-    user_prompt_str=message
-    chat_history=[]
-    max_new_tokens=512
-
-    formated_prompt_list, user_prompt_list = format_user_prompt_with_im_history_and_system_conditioning(
-        current_user_prompt_str=user_prompt_str.strip(),
-        current_image=image,
-        history=chat_history,
-    )
-    '''
-    client_endpoint = API_PATHS[model_selector]
-    client = Client(
-        base_url=client_endpoint,
-        headers={"x-use-cache": "0", "Authorization": f"Bearer {API_TOKEN}"},
-        timeout=240, # Generous time out just in case because we are in greedy. All examples should be computed in less than 30secs with the 80b-instruct.
-    )'''
-
-    # Common parameters to all decoding strategies
-    # This documentation is useful to read: https://huggingface.co/docs/transformers/main/en/generation_strategies
-    generation_args = {
-        "max_new_tokens": max_new_tokens,
-        "repetition_penalty": None,
-        "stop_sequences": EOS_STRINGS,
-        "do_sample": False,
-    }
-
-    if image is None:
-        # Case where there is no image OR the image is passed as `<fake_token_around_image><image:IMAGE_URL><fake_token_around_image>`
-        chat_history.append([prompt_list_to_markdown(user_prompt_list), ''])
-    else:
-        # Case where the image is passed through the Image Box.
-        # Convert the image into base64 for both passing it through the chat history and
-        # displaying the image inside the same bubble as the text.
-        chat_history.append(
-            [
-                f"{prompt_list_to_markdown([image] + user_prompt_list)}",
-                '',
-            ]
-        )
-
-    # Hack - see explanation in `DEFAULT_IMAGES_TMP_PATH_TO_URL`
-    for idx, i in enumerate(formated_prompt_list):
-        if i.startswith(DEFAULT_TEMP_DIR):
-            for k, v in DEFAULT_IMAGES_TMP_PATH_TO_URL.items():
-                if k == i:
-                    formated_prompt_list[idx] = v
-                    break
-
-    query = prompt_list_to_tgi_input(formated_prompt_list)
-    generated_text = client.generate(prompt=query, **generation_args).generated_text
-    if generated_text.endswith("\nUser"):
-        generated_text = generated_text[:-5]
+        "\nAssistant:",
+    ],
+]
 
-    last_turn = chat_history.pop(-1)
-    last_turn[-1] += generated_text
-    chat_history.append(last_turn)
-    return "", None, chat_history
+# --batched mode
+inputs = processor(prompts, add_end_of_utterance_token=False, return_tensors="pt").to(device)
+# --single sample mode
+# inputs = processor(prompts[0], return_tensors="pt").to(device)
 
-with gr.Blocks() as app:
-    chatbot=gr.Chatbot()
-    inp_im = gr.Image()
-    inp_txt = gr.Textbox()
-    
-    sub_btn = gr.Button()
+# Generation args
+exit_condition = processor.tokenizer("<end_of_utterance>", add_special_tokens=False).input_ids
+bad_words_ids = processor.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids
 
-    
-    sub_btn.click(process_example,[inp_txt,inp_im],[inp_txt,inp_im,chatbot])
-app.launch()
+generated_ids = model.generate(**inputs, eos_token_id=exit_condition, bad_words_ids=bad_words_ids, max_length=100)
+generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
+for i, t in enumerate(generated_text):
+    print(f"{i}:\n{t}\n")