# 测试基础模型推理
if __name__ == "__main__":
    # 模型路径
    model_path = "../models/Qwen/Qwen2___5-3B-Instruct"
    # 加载模型和分词器
    model, tokenizer = get_model_tokenizer(model_path)
    # 测试分词结果
    print("测试分词结果：{}".format(tokenizer("文本转词TOKEN")))
    # 获取模板
    template = get_template(
        model.model_meta.template,
        tokenizer,
        default_system="you are a helper assistant",
        max_length=128,
        padding_side="right",
    )
    # 创建输入消息
    message: StdTemplateInputs = StdTemplateInputs(messages=[{"role": "user", "content": "你是谁"}])
    # 编码输入消息
    resp = template.encode(message)
    # 将编码后的输入转换为张量并移动到模型设备
    resp = {key: torch.tensor(value).to(model.device) for key, value in resp.items()}
    # 模型推理
    with torch.no_grad():
        output = model(**resp)
    # 解码输出
    generated_ids = output.logits.argmax(dim=-1)
    generated_text = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
    print("生成的文本：", generated_text)