import gradio as gr
from llama_cpp import Llama
import json
import time
llm = Llama(model_path="./neuralhermes-2.5-mistral-7b.Q5_K_M.gguf",
            n_ctx=32768,
            n_threads=2,
            chat_format="chatml")
llm_for_translate = Llama(model_path="./qwen-1.8b-q5_k_m.gguf",
            n_ctx=1024,
            n_threads=2,
            chat_format="chatml")

chi_eng_dict = []


def get_dict_result(original_text):
    for d in chi_eng_dict:
        if d[0] == original_text:
            return d[1]
        elif d[1] == original_text:
            return d[0]
    return None

def stream_translate_into(message, language='English'):
    return llm_for_translate.create_chat_completion(
        messages=[{"role": "system", "content": f"Translate words into {language}. Regardless the meanning!"},
                    {"role": "user", "content": f"'{message}'"}],
        stream=True,
        stop=['\n\n']
    )

def chat_completion(message, history, system_prompt):
    messages_prompts = [{"role": "system", "content": system_prompt}]
    for human, assistant in history:
        messages_prompts.append({"role": "user", "content": human})
        messages_prompts.append({"role": "assistant", "content": assistant})
    messages_prompts.append({"role": "user", "content": message})
    response = llm.create_chat_completion(
        messages=messages_prompts,
        stream=False
    )
    print(json.dumps(response, ensure_ascii=False, indent=2))
    return response['choices'][0]['message']['content']


def chat_stream_completion(message, history, system_prompt, translate_check):
    messages_prompts = [{"role": "system", "content": system_prompt}]
    if translate_check:
        if len(history) > 0:
            for human, assistant in history:
                human_repl = get_dict_result(human)
                assistant_repl = get_dict_result(assistant)
                if human_repl is None or assistant_repl is None:
                    print(chi_eng_dict)
                    raise gr.Error("历史信息缺少翻译字典，请勿中途修改翻译功能！")
                messages_prompts.append({"role": "user", "content": human_repl})
                messages_prompts.append({"role": "assistant", "content": assistant_repl})
        message_repl = ""
        for chunk in stream_translate_into(message, language='English'):
            if len(chunk['choices'][0]["delta"]) != 0 and "content" in chunk['choices'][0]["delta"]:
                message_repl = message_repl + \
                    chunk['choices'][0]["delta"]["content"]
        chi_eng_dict.append((message, message_repl))
        messages_prompts.append({"role": "user", "content": message_repl})
        print(messages_prompts)
        response = llm.create_chat_completion(
            messages=messages_prompts,
            stream=False,
            stop=['\n\n']
        )
        print(json.dumps(response, ensure_ascii=False, indent=2))
        result = response['choices'][0]['message']['content']
        result_repl = ""
        for chunk in stream_translate_into(result, language='Chinese'):
            if len(chunk['choices'][0]["delta"]) != 0 and "content" in chunk['choices'][0]["delta"]:
                result_repl = result_repl + \
                    chunk['choices'][0]["delta"]["content"]
                yield result_repl
        chi_eng_dict.append((result, result_repl))
    else:
        for human, assistant in history:
            messages_prompts.append({"role": "user", "content": human})
            messages_prompts.append({"role": "assistant", "content": assistant})
        messages_prompts.append({"role": "user", "content": message})
        response = llm.create_chat_completion(
            messages=messages_prompts,
            stream=True
        )
        message_repl = ""
        for chunk in response:
            if len(chunk['choices'][0]["delta"]) != 0 and "content" in chunk['choices'][0]["delta"]:
                message_repl = message_repl + \
                    chunk['choices'][0]["delta"]["content"]
            yield message_repl


gr.ChatInterface(
    chat_stream_completion,
    additional_inputs=[gr.Textbox(
        "You are helpful AI.", label="System Prompt"), gr.Checkbox(label="Translate?")]
).queue().launch(server_name="0.0.0.0")