app.py

import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

# 设置设备，如果有GPU则使用GPU
# 在免费CPU服务器上，这里会是 'cpu'
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")

# --- 修改部分开始 ---
# 只加载Chimera集成模型和分词器
print("Loading Hunyuan-MT-Chimera-7B model and tokenizer...")
tokenizer = AutoTokenizer.from_pretrained("tencent/Hunyuan-MT-Chimera-7B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("tencent/Hunyuan-MT-Chimera-7B", trust_remote_code=True).to(device)
print("Model loaded successfully.")
# --- 修改部分结束 ---


def translate(text_to_translate, source_lang, target_lang):
    """
    使用已加载的Chimera模型进行翻译
    """
    prompt = f"Translate the following text from {source_lang} to {target_lang}:\n{text_to_translate}"
    
    inputs = tokenizer(prompt, return_tensors="pt").to(device)
    
    # 生成翻译结果
    output = model.generate(**inputs, max_new_tokens=256)
    
    # 解码并清理结果
    response = tokenizer.decode(output[0], skip_special_tokens=True)
    
    # 移除prompt部分，只返回翻译结果
    translated_text = response.replace(prompt, "").strip()
    
    return translated_text

# --- 创建Gradio界面 ---
with gr.Blocks() as demo:
    gr.Markdown("# 腾讯混元翻译模型体验Demo")
    gr.Markdown("模型: Hunyuan-MT-Chimera-7B (集成优化版)")
    
    with gr.Row():
        source_language = gr.Textbox(label="源语言", value="Chinese")
        target_language = gr.Textbox(label="目标语言", value="English")

    input_text = gr.Textbox(label="输入文本", lines=5, placeholder="在这里输入需要翻译的文本...")
    output_text = gr.Textbox(label="翻译结果", lines=5)
    
    translate_button = gr.Button("开始翻译")
    
    translate_button.click(
        fn=translate,
        inputs=[input_text, source_language, target_language],
        outputs=output_text
    )

# 启动应用
demo.launch()