app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, StoppingCriteria
import gradio as gr
import spaces
import torch
import numpy as np
import torch
import torchvision.transforms as T
from PIL import Image
from torchvision.transforms.functional import InterpolationMode
from transformers import AutoModel, AutoTokenizer

from threading import Thread
import re
import time 
from PIL import Image
import torch
import spaces
import subprocess
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)

torch.set_default_device('cuda')


IMAGENET_MEAN = (0.485, 0.456, 0.406)
IMAGENET_STD = (0.229, 0.224, 0.225)

def build_transform(input_size):
    MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
    transform = T.Compose([
        T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
        T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
        T.ToTensor(),
        T.Normalize(mean=MEAN, std=STD)
    ])
    return transform

def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    best_ratio_diff = float('inf')
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio
    return best_ratio

def dynamic_preprocess(image, min_num=1, max_num=12, image_size=448, use_thumbnail=False):
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    # calculate the existing image aspect ratio
    target_ratios = set(
        (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
        i * j <= max_num and i * j >= min_num)
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])

    # find the closest aspect ratio to the target
    target_aspect_ratio = find_closest_aspect_ratio(
        aspect_ratio, target_ratios, orig_width, orig_height, image_size)

    # calculate the target width and height
    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    # resize the image
    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = (
            (i % (target_width // image_size)) * image_size,
            (i // (target_width // image_size)) * image_size,
            ((i % (target_width // image_size)) + 1) * image_size,
            ((i // (target_width // image_size)) + 1) * image_size
        )
        # split the image
        split_img = resized_img.crop(box)
        processed_images.append(split_img)
    assert len(processed_images) == blocks
    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)
    return processed_images

def load_image(image_file, input_size=448, max_num=12):
    image = Image.open(image_file).convert('RGB')
    transform = build_transform(input_size=input_size)
    images = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num)
    pixel_values = [transform(image) for image in images]
    pixel_values = torch.stack(pixel_values)
    return pixel_values

model = AutoModel.from_pretrained(
    "5CD-AI/Viet-InternVL2-1B",
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    trust_remote_code=True,
).eval().cuda()
tokenizer = AutoTokenizer.from_pretrained("5CD-AI/Viet-InternVL2-1B", trust_remote_code=True, use_fast=False)


@spaces.GPU
def chat(message, history):
    print(history)
    print(message)
    if len(history) == 0 or len(message["files"]) != 0:
        test_image = message["files"][0]["path"]
    else:
        test_image = history[0][0][0]
        
    pixel_values = load_image(test_image, max_num=12).to(torch.bfloat16).cuda()
    generation_config = dict(max_new_tokens= 1024, do_sample=True, num_beams = 3, repetition_penalty=2.5)
    
    
    
    if len(history) == 0:
        question = '<image>\n'+message["text"]
        response, conv_history = model.chat(tokenizer, pixel_values, question, generation_config, history=None, return_history=True)
    else:
        conv_history = []
        for chat_pair in history:
            if chat_pair[1] is not None:
                if len(conv_history) == 0 and len(message["files"]) == 0:
                    chat_pair[0] = '<image>\n' + chat_pair[0]
                conv_history.append(tuple(chat_pair))
        print(conv_history)
        if len(message["files"]) != 0:
            question = '<image>\n'+message["text"]
        else:
            question = message["text"]
        response, conv_history = model.chat(tokenizer, pixel_values, question, generation_config, history=conv_history, return_history=True)
        
    print(f'User: {question}\nAssistant: {response}')

    buffer = ""
    for new_text in response:
      buffer += new_text
      generated_text_without_prompt = buffer[:]
      time.sleep(0.01)
      yield generated_text_without_prompt

CSS ="""
.contain { display: flex; flex-direction: column; 
  height: 200%; /* Tăng chiều cao của container gấp đôi */
}
#component-0 { height: 100%; }
#component-4 {
  height: 30dvh !important;
  transform-origin: top; /* Đảm bảo rằng phần tử mở rộng từ trên xuống */
  border-style: solid;
  overflow: hidden;
  flex-grow: 1;
  min-width: min(160px, 100%);
  border-width: var(--block-border-width);
}
"""

with gr.Blocks(css=CSS) as demo:
    chatbot = gr.ChatInterface(
        fn=chat,
        description="""Try [Vintern-1B](https://huggingface.co/5CD-AI/Viet-InternVL2-1B) in this demo. Vintern 1B is a multimodal large language model series, featuring models of various sizes. For each size, we release instruction-tuned models optimized for multimodal tasks. Vintern-1B consists of [InternViT-300M-448px](https://huggingface.co/OpenGVLab/InternViT-300M-448px), an MLP projector, and [Qwen2-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2-0.5B-Instruct).""",
        examples=[{"text": "Mô tả hình ảnh.", "files":["./demo_3.jpg"]},
                  {"text": "Trích xuất các thông tin từ ảnh.", "files":["./demo_1.jpg"]}, 
                  {"text": "Mô tả hình ảnh một cách chi tiết.", "files":["./demo_2.jpg"]}],
        title="❄️ Vintern-1B ❄️",
        multimodal=True,
    )
demo.queue().launch()