import os
import platform
import signal
from transformers import AutoTokenizer, AutoModel
import readline


base_model = '../FinGPT/FinGPT_mt_chatglm2-6b-merged'

tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
model = AutoModel.from_pretrained(base_model, trust_remote_code=True, device_map = "auto")


# 多显卡支持，使用下面两行代替上面一行，将num_gpus改为你实际的显卡数量
# from utils import load_model_on_gpus
# model = load_model_on_gpus("THUDM/chatglm2-6b", num_gpus=2)
model = model.eval()

os_name = platform.system()
clear_command = 'cls' if os_name == 'Windows' else 'clear'
stop_stream = False


def build_prompt(history):
    prompt = "欢迎使用 FinGPT-ChatGLM2-6B 模型，输入内容即可进行对话，clear 清空对话历史，stop 终止程序"
    for query, response in history:
        prompt += f"\n\n用户：{query}"
        prompt += f"\n\nFinGPT：{response}"
    return prompt


def signal_handler(signal, frame):
    global stop_stream
    stop_stream = True


def main():
    os.system(clear_command)
    past_key_values, history = None, []
    global stop_stream
    print("欢迎使用由中科曙光智能与计算产业事业部开发的FinGPT-ChatGLM2-6B金融大模型，输入内容即可进行对话，clear 清空对话历史，stop 终止程序")
    while True:
        query = input("\n用户：")
        if query.strip() == "stop":
            break
        if query.strip() == "clear":
            past_key_values, history = None, []
            os.system(clear_command)
            print("欢迎使用由中科曙光智能与计算产业事业部开发的FinGPT-ChatGLM2-6B金融大模型，输入内容即可进行对话，clear 清空对话历史，stop 终止程序")
            continue
        print("\nFinGPT：", end="")
        current_length = 0
        for response, history, past_key_values in model.stream_chat(tokenizer, query, history=history,
                                                                    past_key_values=past_key_values,
                                                                    return_past_key_values=True,
                                                                    temperature=0.8,
                                                                    ):
            if stop_stream:
                stop_stream = False
                break
            else:
                print(response[current_length:], end="", flush=True)
                current_length = len(response)
        print("")


if __name__ == "__main__":
    main()