app.py

import gradio as gr
import torch
import numpy as np
from PIL import Image
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

MODEL_PATH = "THUDM/cogvlm2-video-llama3-chat"
DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
TORCH_TYPE = torch.bfloat16 if torch.cuda.is_available() and torch.cuda.get_device_capability()[0] >= 8 else torch.float16

def load_model():
    """Loads the pre-trained model and tokenizer with quantization configurations."""
    quantization_config = BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_compute_dtype=TORCH_TYPE,
        bnb_4bit_use_double_quant=True,
        bnb_4bit_quant_type="nf4"
    )
    
    tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
    model = AutoModelForCausalLM.from_pretrained(
        MODEL_PATH,
        torch_dtype=TORCH_TYPE,
        trust_remote_code=True,
        quantization_config=quantization_config,
        device_map="auto"
    ).eval()
    
    return model, tokenizer

def predict_image(prompt, image, temperature, model, tokenizer):
    """Generates predictions based on the image and textual prompt."""
    image = image.convert("RGB")  # Ensure image is in RGB format
    
    # Convert image to model-expected format
    inputs = model.build_conversation_input_ids(
        tokenizer=tokenizer,
        query=prompt,
        images=[image],
        history=[],
        template_version='chat'
    )
    
    inputs = {
        'input_ids': inputs['input_ids'].unsqueeze(0).to(DEVICE),
        'token_type_ids': inputs['token_type_ids'].unsqueeze(0).to(DEVICE),
        'attention_mask': inputs['attention_mask'].unsqueeze(0).to(DEVICE),
        'images': [[inputs['images'][0].to(DEVICE).to(TORCH_TYPE)]],
    }
    
    gen_kwargs = {
        "max_new_tokens": 512,
        "pad_token_id": 128002,
        "top_k": 1,
        "do_sample": False,
        "top_p": 0.1,
        "temperature": temperature,
    }
    
    with torch.no_grad():
        outputs = model.generate(**inputs, **gen_kwargs)
        outputs = outputs[:, inputs['input_ids'].shape[1]:]
        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    return response

model, tokenizer = load_model()

def inference(image):
    """Generates a description of the input image."""
    try:
        if not image:
            return "Please upload an image first."
        
        prompt = "Describe the image and the components observed in the given input image."
        temperature = 0.3
        response = predict_image(prompt, image, temperature, model, tokenizer)
        
        return response
    except Exception as e:
        return f"An error occurred during analysis: {str(e)}"

def create_interface():
    """Creates the Gradio interface for Image Description System."""
    with gr.Blocks() as demo:
        gr.Markdown("""
        # Image Description System
        Upload an image, and the system will describe the image and its components.
        """)
        
        with gr.Row():
            with gr.Column():
                image_input = gr.Image(label="Upload Image", type="pil")
                analyze_btn = gr.Button("Describe Image", variant="primary")
            
            with gr.Column():
                output = gr.Textbox(label="Image Description", lines=10)
        
        analyze_btn.click(
            fn=inference,
            inputs=[image_input],
            outputs=[output]
        )
    
    return demo

if __name__ == "__main__":
    demo = create_interface()
    demo.queue().launch(share=True)