app.py

import gradio as gr
import os
from threading import Thread
from queue import Queue
import time
import cv2
import datetime
import torch
import spaces
import numpy as np
import json
import hashlib
import PIL
from typing import Iterator

from llava import conversation as conversation_lib
from llava.constants import DEFAULT_IMAGE_TOKEN
from llava.constants import (
    IMAGE_TOKEN_INDEX,
    DEFAULT_IMAGE_TOKEN,
    DEFAULT_IM_START_TOKEN,
    DEFAULT_IM_END_TOKEN,
)
from llava.conversation import conv_templates, SeparatorStyle
from llava.model.builder import load_pretrained_model
from llava.utils import disable_torch_init
from llava.mm_utils import (
    tokenizer_image_token,
    get_model_name_from_path,
    KeywordsStoppingCriteria,
)

from serve_constants import html_header

import requests
from PIL import Image
from io import BytesIO
from transformers import TextIteratorStreamer

external_log_dir = "./logs"
LOGDIR = external_log_dir

def install_gradio_4_35_0():
    current_version = gr.__version__
    if current_version != "4.35.0":
        print(f"Current Gradio version: {current_version}")
        print("Installing Gradio 4.35.0...")
        subprocess.check_call([sys.executable, "-m", "pip", "install", "gradio==4.35.0", "--force-reinstall"])
        print("Gradio 4.35.0 installed successfully.")
    else:
        print("Gradio 4.35.0 is already installed.")

install_gradio_4_35_0()

print(f"Gradio version: {gr.__version__}")

def get_conv_log_filename():
    t = datetime.datetime.now()
    name = os.path.join(LOGDIR, f"{t.year}-{t.month:02d}-{t.day:02d}-user_conv.json")
    return name

class InferenceDemo(object):
    def __init__(
        self, args, model_path, tokenizer, model, image_processor, context_len
    ) -> None:
        disable_torch_init()

        self.tokenizer = tokenizer
        self.model = model
        self.image_processor = image_processor
        self.context_len = context_len
        
        model_name = get_model_name_from_path(model_path)

        if "llama-2" in model_name.lower():
            conv_mode = "llava_llama_2"
        elif "v1" in model_name.lower():
            conv_mode = "llava_v1"
        elif "mpt" in model_name.lower():
            conv_mode = "mpt"
        elif "qwen" in model_name.lower():
            conv_mode = "qwen_1_5"
        elif "pangea" in model_name.lower():
            conv_mode = "qwen_1_5"
        else:
            conv_mode = "llava_v0"

        if args.conv_mode is not None and conv_mode != args.conv_mode:
            print(
                "[WARNING] the auto inferred conversation mode is {}, while `--conv-mode` is {}, using {}".format(
                    conv_mode, args.conv_mode, args.conv_mode
                )
            )
        else:
            args.conv_mode = conv_mode
        
        self.conv_mode = conv_mode
        self.conversation = conv_templates[args.conv_mode].copy()
        self.num_frames = args.num_frames

def process_stream(streamer: TextIteratorStreamer, history: list, q: Queue):
    """Process the output stream and put partial text into a queue"""
    try:
        current_message = ""
        for new_text in streamer:
            current_message += new_text
            history[-1][1] = current_message
            q.put(history.copy())
            time.sleep(0.02)  # Add a small delay to prevent overloading
    except Exception as e:
        print(f"Error in process_stream: {e}")
    finally:
        q.put(None)  # Signal that we're done

def stream_output(history: list, q: Queue) -> Iterator[list]:
    """Yield updated history as it comes through the queue"""
    while True:
        val = q.get()
        if val is None:
            break
        yield val
        q.task_done()

def is_valid_video_filename(name):
    video_extensions = ["avi", "mp4", "mov", "mkv", "flv", "wmv", "mjpeg"]
    ext = name.split(".")[-1].lower()
    return ext in video_extensions

def is_valid_image_filename(name):
    image_extensions = ["jpg", "jpeg", "png", "bmp", "gif", "tiff", "webp", "heic", "heif", "jfif", "svg", "eps", "raw"]
    ext = name.split(".")[-1].lower()
    return ext in image_extensions

def sample_frames(video_file, num_frames):
    video = cv2.VideoCapture(video_file)
    total_frames = int(video.get(cv2.CAP_PROP_FRAME_COUNT))
    interval = total_frames // num_frames
    frames = []
    for i in range(total_frames):
        ret, frame = video.read()
        if not ret:
            continue
        pil_img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
        if i % interval == 0:
            frames.append(pil_img)
    video.release()
    return frames

def load_image(image_file):
    if image_file.startswith(("http://", "https://")):
        response = requests.get(image_file)
        if response.status_code == 200:
            image = Image.open(BytesIO(response.content)).convert("RGB")
        else:
            print("Failed to load the image")
            return None
    else:
        print("Load image from local file:", image_file)
        image = Image.open(image_file).convert("RGB")
    return image

def clear_history(history):
    global our_chatbot
    our_chatbot.conversation = conv_templates[our_chatbot.conv_mode].copy()
    return None

def add_message(history, message):
    global our_chatbot
    if len(history) == 0:
        our_chatbot = InferenceDemo(
            args, model_path, tokenizer, model, image_processor, context_len
        )

    for x in message["files"]:
        history.append(((x,), None))
    if message["text"] is not None:
        history.append((message["text"], None))
    return history, gr.MultimodalTextbox(value=None, interactive=False)

@spaces.GPU
def bot(history):
    global start_tstamp, finish_tstamp
    
    start_tstamp = time.time()
    text = history[-1][0]
    images_this_term = []
    num_new_images = 0
    
    for i, message in enumerate(history[:-1]):
        if isinstance(message[0], tuple):
            images_this_term.append(message[0][0])
            if is_valid_video_filename(message[0][0]):
                raise ValueError("Video is not supported")
            elif is_valid_image_filename(message[0][0]):
                num_new_images += 1
            else:
                raise ValueError("Invalid image file")
        else:
            num_new_images = 0

    assert len(images_this_term) > 0, "Must have an image"
    
    image_list = []
    for f in images_this_term:
        if is_valid_video_filename(f):
            image_list += sample_frames(f, our_chatbot.num_frames)
        elif is_valid_image_filename(f):
            image_list.append(load_image(f))
        else:
            raise ValueError("Invalid image file")
    
    image_tensor = [
        our_chatbot.image_processor.preprocess(f, return_tensors="pt")["pixel_values"][0]
        .half()
        .to(our_chatbot.model.device)
        for f in image_list
    ]
    
    # Process image hashes
    all_image_hash = []
    for image_path in images_this_term:
        with open(image_path, "rb") as image_file:
            image_data = image_file.read()
            image_hash = hashlib.md5(image_data).hexdigest()
            all_image_hash.append(image_hash)
            image = PIL.Image.open(image_path).convert("RGB")
            t = datetime.datetime.now()
            filename = os.path.join(
                LOGDIR,
                "serve_images",
                f"{t.year}-{t.month:02d}-{t.day:02d}",
                f"{image_hash}.jpg",
            )
            if not os.path.isfile(filename):
                os.makedirs(os.path.dirname(filename), exist_ok=True)
                image.save(filename)

    image_tensor = torch.stack(image_tensor)
    image_token = DEFAULT_IMAGE_TOKEN * num_new_images
    inp = image_token + "\n" + text
    
    our_chatbot.conversation.append_message(our_chatbot.conversation.roles[0], inp)
    our_chatbot.conversation.append_message(our_chatbot.conversation.roles[1], None)
    prompt = our_chatbot.conversation.get_prompt()

    input_ids = (
        tokenizer_image_token(
            prompt, our_chatbot.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
        )
        .unsqueeze(0)
        .to(our_chatbot.model.device)
    )

    stop_str = (
        our_chatbot.conversation.sep
        if our_chatbot.conversation.sep_style != SeparatorStyle.TWO
        else our_chatbot.conversation.sep2
    )
    keywords = [stop_str]
    stopping_criteria = KeywordsStoppingCriteria(
        keywords, our_chatbot.tokenizer, input_ids
    )

    # Set up streaming
    q = Queue()
    streamer = TextIteratorStreamer(
        our_chatbot.tokenizer,
        skip_prompt=True,
        skip_special_tokens=True
    )

    # Start generation in a separate thread
    thread = Thread(
        target=process_stream,
        args=(streamer, history, q)
    )
    thread.start()

    # Start the generation
    with torch.inference_mode():
        output_ids = our_chatbot.model.generate(
            input_ids,
            images=image_tensor,
            do_sample=True,
            temperature=0.2,
            max_new_tokens=1024,
            streamer=streamer,
            use_cache=True,
            stopping_criteria=[stopping_criteria],
        )

    finish_tstamp = time.time()
    
    # Log conversation
    with open(get_conv_log_filename(), "a") as fout:
        data = {
            "tstamp": round(finish_tstamp, 4),
            "type": "chat",
            "model": "Pangea-7b",
            "start": round(start_tstamp, 4),
            "finish": round(finish_tstamp, 4),
            "state": history,
            "images": all_image_hash,
        }
        fout.write(json.dumps(data) + "\n")

    # Return a generator that will yield updated history
    return stream_output(history, q)

with gr.Blocks(css=".message-wrap.svelte-1lcyrx4>div.svelte-1lcyrx4 img {min-width: 40px}") as demo:
    gr.HTML(html_header)
    
    with gr.Column():
        with gr.Row():
            chatbot = gr.Chatbot([], elem_id="Pangea", bubble_full_width=False, height=750)

        with gr.Row():
            upvote_btn = gr.Button(value="👍  Upvote", interactive=True)
            downvote_btn = gr.Button(value="👎  Downvote", interactive=True)
            flag_btn = gr.Button(value="⚠️  Flag", interactive=True)
            regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=True)
            clear_btn = gr.Button(value="🗑️  Clear history", interactive=True)

        chat_input = gr.MultimodalTextbox(
            interactive=True,
            file_types=["image"],
            placeholder="Enter message or upload file...",
            show_label=False,
            submit_btn="🚀"
        )

        cur_dir = os.path.dirname(os.path.abspath(__file__))
        gr.Examples(
            examples_per_page=20,
            examples=[
                [
                        {
                            "files": [
                                f"{cur_dir}/examples/user_example_07.jpg",
                            ],
                            "text": "那要我问问你，你这个是什么🐱？",
                        },
                    ],
                    [
                        {
                            "files": [
                                f"{cur_dir}/examples/user_example_05.jpg",
                            ],
                            "text": "この猫の目の大きさは、どのような理由で他の猫と比べて特に大きく見えますか？",
                        },
                    ],
                    [
                        {
                            "files": [
                                f"{cur_dir}/examples/172197131626056_P7966202.png",
                            ],
                            "text": "Why this image funny?",
                        },
                    ],
            ],
            inputs=[chat_input],
            label="Image",
        )

    chat_msg = chat_input.submit(
        add_message, 
        [chatbot, chat_input], 
        [chatbot, chat_input],
        queue=False
    ).then(
        bot,
        chatbot,
        chatbot,
        api_name="bot_response"
    ).then(
        lambda: gr.MultimodalTextbox(interactive=True), 
        None, 
        [chat_input]
    )

    clear_btn.click(
        fn=clear_history, 
        inputs=[chatbot], 
        outputs=[chatbot], 
        api_name="clear_all",
        queue=False
    )

    regenerate_btn.click(
        fn=lambda history: history[:-1],
        inputs=[chatbot],
        outputs=[chatbot],
        queue=False
    ).then(
        bot,
        chatbot,
        chatbot
    )

demo.queue(concurrency_count=5)
    
if __name__ == "__main__":
    import argparse

    argparser = argparse.ArgumentParser()
    argparser.add_argument("--server_name", default="0.0.0.0", type=str)
    argparser.add_argument("--port", default="6123", type=str)
    argparser.add_argument(
        "--model_path", default="neulab/Pangea-7B", type=str
    )
    # argparser.add_argument("--model-path", type=str, default="facebook/opt-350m")
    argparser.add_argument("--model-base", type=str, default=None)
    argparser.add_argument("--num-gpus", type=int, default=1)
    argparser.add_argument("--conv-mode", type=str, default=None)
    argparser.add_argument("--temperature", type=float, default=0.7)
    argparser.add_argument("--max-new-tokens", type=int, default=4096)
    argparser.add_argument("--num_frames", type=int, default=16)
    argparser.add_argument("--load-8bit", action="store_true")
    argparser.add_argument("--load-4bit", action="store_true")
    argparser.add_argument("--debug", action="store_true")

    args = argparser.parse_args()

    model_path = args.model_path
    filt_invalid = "cut"
    model_name = get_model_name_from_path(args.model_path)
    tokenizer, model, image_processor, context_len = load_pretrained_model(args.model_path, args.model_base, model_name, args.load_8bit, args.load_4bit)
    model=model.to(torch.device('cuda'))
    our_chatbot = None
    demo.launch()