app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
from PIL import Image
import base64
from io import BytesIO

# 加载模型和分词器
model_name = "openbmb/MiniCPM-Llama3-V-2_5-int4"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)

def encode_image(image):
    buffered = BytesIO()
    image.save(buffered, format="PNG")
    return base64.b64encode(buffered.getvalue()).decode('utf-8')

def generate_text(prompt, max_length=100):
    inputs = tokenizer(prompt, return_tensors="pt")
    with torch.no_grad():
        outputs = model.generate(**inputs, max_length=max_length, num_return_sequences=1)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

def predict(image, prompt):
    if image is not None:
        # 确保image是PIL Image对象
        if isinstance(image, str):
            image = Image.open(image)
        
        # 编码图像
        encoded_image = encode_image(image)
        
        # 准备输入
        full_prompt = f"<image>{encoded_image}</image>\n{prompt if prompt else 'Describe this image.'}"
        
        # 生成文本
        result = generate_text(full_prompt)
        
        return f"Model response: {result}\n\nUser prompt: {prompt}"
    else:
        return "No image uploaded. " + (f"You asked: {prompt}" if prompt else "Please upload an image and optionally provide a prompt.")

demo = gr.Interface(
    predict,
    inputs=[
        gr.Image(type="pil", label="Upload Image"),
        gr.Textbox(label="Prompt (optional)")
    ],
    outputs=gr.Textbox(label="Result"),
    title="Image Analysis with MiniCPM-Llama3-V-2_5-int4",
    description="Upload an image and optionally provide a prompt for analysis."
)

if __name__ == "__main__":
    demo.launch()