app.py

import os
import subprocess

# Install flash attention
# subprocess.run(
#     "pip install flash-attn --no-build-isolation",
#     env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
#     shell=True,
# )


import copy
import spaces
import time
import torch

from threading import Thread
from typing import List, Dict, Union
import urllib
from PIL import Image
import io
import datasets

import bitsandbytes

import gradio as gr
from transformers import AutoProcessor, TextIteratorStreamer
from transformers import Idefics2ForConditionalGeneration

import torch
from peft import LoraConfig
from transformers import AutoProcessor, BitsAndBytesConfig, IdeficsForVisionText2Text


DEVICE = torch.device("cuda")

USE_LORA = False
USE_QLORA = True

if USE_QLORA or USE_LORA:
    lora_config = LoraConfig(
        r=8,
        lora_alpha=8,
        lora_dropout=0.1,
        target_modules='.*(text_model|modality_projection|perceiver_resampler).*(down_proj|gate_proj|up_proj|k_proj|q_proj|v_proj|o_proj).*$',
        use_dora=False if USE_QLORA else True,
        init_lora_weights="gaussian"
    )
    if USE_QLORA:
        bnb_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_quant_type="nf4",
            bnb_4bit_compute_dtype=torch.float16
        )

    MODELS = {
    "idefics2-8b-vqarad-delta": Idefics2ForConditionalGeneration.from_pretrained(
        "jihadzakki/idefics2-8b-vqarad-delta",
        torch_dtype=torch.float16,
        quantization_config=bnb_config if USE_QLORA else None,
    )
    }

PROCESSOR = AutoProcessor.from_pretrained(
    "HuggingFaceM4/idefics2-8b",
)

SYSTEM_PROMPT = [
    {
        "role": "system",
        "content": [
            {
                "type": "text",
                "text": "The following is a conversation between Idefics2, a highly knowledgeable and intelligent visual AI assistant created by Hugging Face, referred to as 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 them, but it cannot generate images. 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.",
            },
        ],
    },
    {
        "role": "assistant",
        "content": [
            {
                "type": "text",
                "text": "Hello, I'm Idefics2, Huggingface's latest multimodal assistant. How can I help you?",
            },
        ],
    }
]

# Chatbot utils
def turn_is_pure_media(turn):
    return turn[1] is None


def load_image_from_url(url):
    with urllib.request.urlopen(url) as response:
        image_data = response.read()
        image_stream = io.BytesIO(image_data)
        image = Image.open(image_stream)
        return image


def img_to_bytes(image_path):
    image = Image.open(image_path).convert(mode='RGB')
    buffer = io.BytesIO()
    image.save(buffer, format="JPEG")
    img_bytes = buffer.getvalue()
    image.close()
    return img_bytes


def format_user_prompt_with_im_history_and_system_conditioning(
    user_prompt, chat_history
) -> List[Dict[str, Union[List, str]]]:
    """
    Produces the resulting list that needs to go inside the processor.
    It handles the potential image(s), the history and the system conditionning.
    """
    resulting_messages = copy.deepcopy(SYSTEM_PROMPT)
    resulting_images = []
    for resulting_message in resulting_messages:
        if resulting_message["role"] == "user":
            for content in resulting_message["content"]:
                if content["type"] == "image":
                    resulting_images.append(load_image_from_url(content["image"]))

    # Format history
    for turn in chat_history:
        if not resulting_messages or (
            resulting_messages and resulting_messages[-1]["role"] != "user"
        ):
            resulting_messages.append(
                {
                    "role": "user",
                    "content": [],
                }
            )

        if turn_is_pure_media(turn):
            media = turn[0][0]
            resulting_messages[-1]["content"].append({"type": "image"})
            resulting_images.append(Image.open(media))
        else:
            user_utterance, assistant_utterance = turn
            resulting_messages[-1]["content"].append(
                {"type": "text", "text": user_utterance.strip()}
            )
            resulting_messages.append(
                {
                    "role": "assistant",
                    "content": [{"type": "text", "text": user_utterance.strip()}],
                }
            )

    # Format current input
    if not user_prompt["files"]:
        resulting_messages.append(
            {
                "role": "user",
                "content": [{"type": "text", "text": user_prompt["text"]}],
            }
        )
    else:
        # Choosing to put the image first (i.e. before the text), but this is an arbiratrary choice.
        resulting_messages.append(
            {
                "role": "user",
                "content": [{"type": "image"}] * len(user_prompt["files"])
                + [{"type": "text", "text": user_prompt["text"]}],
            }
        )
        resulting_images.extend([Image.open(path) for path in user_prompt["files"]])

    return resulting_messages, resulting_images


def extract_images_from_msg_list(msg_list):
    all_images = []
    for msg in msg_list:
        for c_ in msg["content"]:
            if isinstance(c_, Image.Image):
                all_images.append(c_)
    return all_images


# @spaces.GPU(duration=180)
def model_inference(
    user_prompt,
    chat_history,
    model_selector,
    decoding_strategy,
    temperature,
    max_new_tokens,
    repetition_penalty,
    top_p,
):
    if user_prompt["text"].strip() == "" and not user_prompt["files"]:
        gr.Error("Please input a query and optionally image(s).")

    if user_prompt["text"].strip() == "" and user_prompt["files"]:
        gr.Error("Please input a text query along the image(s).")

    streamer = TextIteratorStreamer(
        PROCESSOR.tokenizer,
        skip_prompt=True,
        timeout=5.0,
    )

    # 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": repetition_penalty,
        "streamer": streamer,
    }

    assert decoding_strategy in [
        "Greedy",
        "Top P Sampling",
    ]
    if decoding_strategy == "Greedy":
        generation_args["do_sample"] = False
    elif decoding_strategy == "Top P Sampling":
        generation_args["temperature"] = temperature
        generation_args["do_sample"] = True
        generation_args["top_p"] = top_p

    # Creating model inputs
    (
        resulting_text,
        resulting_images,
    ) = format_user_prompt_with_im_history_and_system_conditioning(
        user_prompt=user_prompt,
        chat_history=chat_history,
    )
    prompt = PROCESSOR.apply_chat_template(resulting_text, add_generation_prompt=True)
    inputs = PROCESSOR(
        text=prompt,
        images=resulting_images if resulting_images else None,
        return_tensors="pt",
    )
    inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
    generation_args.update(inputs)

    # The streaming generation mode
    thread = Thread(
        target=MODELS[model_selector].generate,
        kwargs=generation_args,
    )
    thread.start()

    print("Start generating")
    acc_text = ""
    for text_token in streamer:
        time.sleep(0.04)
        acc_text += text_token
        if acc_text.endswith("<end_of_utterance>"):
            acc_text = acc_text[:-18]
        yield acc_text
    print("Success - generated the following text:", acc_text)
    print("-----")


FEATURES = datasets.Features(
    {
        "model_selector": datasets.Value("string"),
        "images": datasets.Sequence(datasets.Image(decode=True)),
        "conversation": datasets.Sequence({"User": datasets.Value("string"), "Assistant": datasets.Value("string")}),
        "decoding_strategy": datasets.Value("string"),
        "temperature": datasets.Value("float32"),
        "max_new_tokens": datasets.Value("int32"),
        "repetition_penalty": datasets.Value("float32"),
        "top_p": datasets.Value("int32"),
        }
    )


# Hyper-parameters for generation
max_new_tokens = gr.Slider(
    minimum=8,
    maximum=1024,
    value=512,
    step=1,
    interactive=True,
    label="Maximum number of new tokens to generate",
)
repetition_penalty = gr.Slider(
    minimum=0.01,
    maximum=5.0,
    value=1.1,
    step=0.01,
    interactive=True,
    label="Repetition penalty",
    info="1.0 is equivalent to no penalty",
)
decoding_strategy = gr.Radio(
    [
        "Greedy",
        "Top P Sampling",
    ],
    value="Greedy",
    label="Decoding strategy",
    interactive=True,
    info="Higher values is equivalent to sampling more low-probability tokens.",
)
temperature = gr.Slider(
    minimum=0.0,
    maximum=5.0,
    value=0.4,
    step=0.1,
    visible=False,
    interactive=True,
    label="Sampling temperature",
    info="Higher values will produce more diverse outputs.",
)
top_p = gr.Slider(
    minimum=0.01,
    maximum=0.99,
    value=0.8,
    step=0.01,
    visible=False,
    interactive=True,
    label="Top P",
    info="Higher values is equivalent to sampling more low-probability tokens.",
)


chatbot = gr.Chatbot(
    label="idefics2-8b-vqarad-delta",
    # avatar_images=[None, BOT_AVATAR],
    height=450,
)

with gr.Blocks(
    fill_height=True,
    css=""".gradio-container .avatar-container {height: 40px width: 40px !important;} #duplicate-button {margin: auto; color: white; background: #f1a139; border-radius: 100vh; margin-top: 2px; margin-bottom: 2px;}""",
) as demo:

    gr.Markdown("# 🐶 Hugging Face Idefics2 8B Chatty")
    gr.Markdown("In this demo you'll be able to chat with [Idefics2-8B-chatty](https://huggingface.co/HuggingFaceM4/idefics2-8b-chatty), a variant of [Idefics2-8B](https://huggingface.co/HuggingFaceM4/idefics2-8b-chatty) further fine-tuned on chat datasets.")
    gr.Markdown("If you want to learn more about Idefics2 and its variants, you can check our [blog post](https://huggingface.co/blog/idefics2).")
    gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")
    # model selector should be set to `visbile=False` ultimately
    with gr.Row(elem_id="model_selector_row"):
        model_selector = gr.Dropdown(
            choices=MODELS.keys(),
            value=list(MODELS.keys())[0],
            interactive=True,
            show_label=False,
            container=False,
            label="Model",
            visible=False,
        )

    decoding_strategy.change(
        fn=lambda selection: gr.Slider(
            visible=(
                selection
                in [
                    "contrastive_sampling",
                    "beam_sampling",
                    "Top P Sampling",
                    "sampling_top_k",
                ]
            )
        ),
        inputs=decoding_strategy,
        outputs=temperature,
    )
    decoding_strategy.change(
        fn=lambda selection: gr.Slider(visible=(selection in ["Top P Sampling"])),
        inputs=decoding_strategy,
        outputs=top_p,
    )

    gr.ChatInterface(
        fn=model_inference,
        chatbot=chatbot,
        # examples=EXAMPLES,
        multimodal=True,
        cache_examples=False,
        additional_inputs=[
            model_selector,
            decoding_strategy,
            temperature,
            max_new_tokens,
            repetition_penalty,
            top_p,
        ],
    )

demo.launch(share=True, debug=True)