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

# 加载模型和 tokenizer
model_name = "LilithHu/mbert-manipulative-detector"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# 设置为评估模式
model.eval()

# 设置运行设备
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# 标签名
labels = ["Non-manipulative / 非操纵性", "Manipulative / 操纵性"]

# 推理函数
def classify(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True).to(device)
    with torch.no_grad():
    outputs = model(**inputs)
    probs = torch.softmax(outputs.logits, dim=1)[0]  # 取第一个样本的概率向量
    probs = torch.clamp(probs, max=0.95)  # 限制最大置信度为 95%
    result = "🧠 预测 / Prediction:\n"
    for i, label in enumerate(labels):
        percent = round(probs[i].item() * 100, 2)
        result += f"{label}: {percent}%\n"
    return result


# Gradio 界面
interface = gr.Interface(
    fn=classify,
    inputs=gr.Textbox(lines=4, placeholder="Enter text in English or Chinese... / 输入中文或英文句子"),
    outputs="text",
    title="🔍 Manipulative Language Detector / 操纵性语言识别器",
    description="🧪 输入英文或中文句子，系统将判断其是否包含操纵性语言。\nEnter a sentence in English or Chinese to detect if it's manipulative.",
)

interface.launch()