'''
https://github.com/openai/whisper
pip install -U openai-whisper
'''
import whisper

class WhisperASR:
    def __init__(self, model_path):
        self.LANGUAGES = {
            "en": "english",
            "zh": "chinese",
        }
        self.model = whisper.load_model(model_path)
        
    def transcribe(self, audio_file):
        result = self.model.transcribe(audio_file)
        return result["text"]

'''
Reference: https://github.com/alibaba-damo-academy/FunASR
pip install funasr
pip install modelscope
pip install -U rotary_embedding_torch
'''
try:
    from funasr import AutoModel
except:
    print("如果想使用FunASR，请先安装funasr，若使用Whisper，请忽略此条信息")   

class FunASR:
    def __init__(self) -> None:
        self.model = AutoModel(model="paraformer-zh", model_revision="v2.0.4",
                vad_model="fsmn-vad", vad_model_revision="v2.0.4",
                punc_model="ct-punc-c", punc_model_revision="v2.0.4",
                # spk_model="cam  ", spk_model_revision="v2.0.2",
                )

    def transcribe(self, audio_file):
        res = self.model.generate(input=audio_file, 
            batch_size_s=300)
        print(res)
        return res[0]['text']

class EdgeTTS:
    def __init__(self, list_voices = False, proxy = None) -> None:
        voices = list_voices_fn(proxy=proxy)
        self.SUPPORTED_VOICE = [item['ShortName'] for item in voices]
        self.SUPPORTED_VOICE.sort(reverse=True)
        if list_voices:
            print(", ".join(self.SUPPORTED_VOICE))

    def preprocess(self, rate, volume, pitch):
        if rate >= 0:
            rate = f' {rate}%'
        else:
            rate = f'{rate}%'
        if pitch >= 0:
            pitch = f' {pitch}Hz'
        else:
            pitch = f'{pitch}Hz'
        volume = 100 - volume
        volume = f'-{volume}%'
        return rate, volume, pitch

    def predict(self,TEXT, VOICE, RATE, VOLUME, PITCH, OUTPUT_FILE='result.wav', OUTPUT_SUBS='result.vtt', words_in_cue = 8):
        async def amain() -> None:
            """Main function"""
            rate, volume, pitch = self.preprocess(rate = RATE, volume = VOLUME, pitch = PITCH)
            communicate = Communicate(TEXT, VOICE, rate = rate, volume = volume, pitch = pitch)
            subs: SubMaker = SubMaker()
            sub_file: Union[TextIOWrapper, TextIO] = (
                open(OUTPUT_SUBS, "w", encoding="utf-8")
            )
            async for chunk in communicate.stream():
                if chunk["type"] == "audio":
                    # audio_file.write(chunk["data"])
                    pass
                elif chunk["type"] == "WordBoundary":
                    # print((chunk["offset"], chunk["duration"]), chunk["text"])
                    subs.create_sub((chunk["offset"], chunk["duration"]), chunk["text"])
            sub_file.write(subs.generate_subs(words_in_cue))
            await communicate.save(OUTPUT_FILE)
            
        
        # loop = asyncio.get_event_loop_policy().get_event_loop()
        # try:
        #     loop.run_until_complete(amain())
        # finally:
        #     loop.close()
        asyncio.run(amain())
        with open(OUTPUT_SUBS, 'r', encoding='utf-8') as file:
            vtt_lines = file.readlines()

        # 去掉每一行文字中的空格
        vtt_lines_without_spaces = [line.replace(" ", "") if "-->" not in line else line for line in vtt_lines]
        # print(vtt_lines_without_spaces)
        with open(OUTPUT_SUBS, 'w', encoding='utf-8') as output_file:
            output_file.writelines(vtt_lines_without_spaces)
        return OUTPUT_FILE, OUTPUT_SUBS

pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu118
# 安装对应的依赖
pip install -r VITS/requirements_gptsovits.txt

# 启动如下的WebUI界面
python VITS/app.py 

class Wav2Lip:
    def __init__(self, path = 'checkpoints/wav2lip.pth'):
        self.fps = 25
        self.resize_factor = 1
        self.mel_step_size = 16
        self.static = False
        self.img_size = 96
        self.face_det_batch_size = 2
        self.box = [-1, -1, -1, -1]
        self.pads = [0, 10, 0, 0]
        self.nosmooth = False
        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
        self.model = self.load_model(path)

    def load_model(self, checkpoint_path):
        model = wav2lip_mdoel()
        print("Load checkpoint from: {}".format(checkpoint_path))
        if self.device == 'cuda':
            checkpoint = torch.load(checkpoint_path)
        else:
            checkpoint = torch.load(checkpoint_path,
                                    map_location=lambda storage, loc: storage)
        s = checkpoint["state_dict"]
        new_s = {}
        for k, v in s.items():
            new_s[k.replace('module.', '')] = v
        model.load_state_dict(new_s)

        model = model.to(self.device)
        return model.eval()

from fastapi import FastAPI, Request
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
import uvicorn
import json
import datetime
import torch
from configs import model_path, api_port
# 设置设备参数
DEVICE = "cuda"  # 使用CUDA
DEVICE_ID = "0"  # CUDA设备ID，如果未设置则为空
CUDA_DEVICE = f"{DEVICE}:{DEVICE_ID}" if DEVICE_ID else DEVICE  # 组合CUDA设备信息

# 清理GPU内存函数
def torch_gc():
    if torch.cuda.is_available():  # 检查是否可用CUDA
        with torch.cuda.device(CUDA_DEVICE):  # 指定CUDA设备
            torch.cuda.empty_cache()  # 清空CUDA缓存
            torch.cuda.ipc_collect()  # 收集CUDA内存碎片

# 创建FastAPI应用
app = FastAPI()

# 处理POST请求的端点
@app.post("/")
async def create_item(request: Request):
    global model, tokenizer  # 声明全局变量以便在函数内部使用模型和分词器
    json_post_raw = await request.json()  # 获取POST请求的JSON数据
    json_post = json.dumps(json_post_raw)  # 将JSON数据转换为字符串
    json_post_list = json.loads(json_post)  # 将字符串转换为Python对象
    prompt = json_post_list.get('prompt')  # 获取请求中的提示
    history = json_post_list.get('history')  # 获取请求中的历史记录
    max_length = json_post_list.get('max_length')  # 获取请求中的最大长度
    top_p = json_post_list.get('top_p')  # 获取请求中的top_p参数
    temperature = json_post_list.get('temperature')  # 获取请求中的温度参数
    
    # 调用模型进行对话生成
    prompt = f"请用少于25个字回答以下问题 ### Instruction:{prompt}  ### Response:"
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda:0")
    generate_ids = model.generate(inputs.input_ids, 
                                  max_new_tokens=max_length if max_length else 2048,
                                  do_sample=True, 
                                  top_k=20,
                                  top_p=top_p,
                                  temperature=temperature if temperature else 0.84,
                                  repetition_penalty=1.15, eos_token_id=2, bos_token_id=1,pad_token_id=0)
    response = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
    response = response.split("### Response:")[-1]
    now = datetime.datetime.now()  # 获取当前时间
    time = now.strftime("%Y-%m-%d %H:%M:%S")  # 格式化时间为字符串
    # 构建响应JSON
    answer = {
        "response": response,
        # "history": history,
        "status": 200,
        "time": time
    }
    # 构建日志信息
    log = "["   time   "] "   '", prompt:"'   prompt   '", response:"'   repr(response)   '"'
    print(log)  # 打印日志
    torch_gc()  # 执行GPU内存清理
    return answer  # 返回响应

# 主函数入口
if __name__ == '__main__':
    # 加载预训练的分词器和模型
    model = AutoModelForCausalLM.from_pretrained(model_path, device_map="cuda:0",
                                                    torch_dtype=torch.bfloat16, trust_remote_code=True)
    tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False, trust_remote_code=True)
    model.eval()  # 设置模型为评估模式
    # 启动FastAPI应用
    uvicorn.run(app, host='0.0.0.0', port=api_port, workers=1)  # 在指定端口和主机上启动应用

import requests
import json

def get_completion(prompt):
    headers = {'Content-Type': 'application/json'}
    data = {"prompt": prompt}
    response = requests.post(url='http://127.0.0.1:7871', headers=headers, data=json.dumps(data))
    return response.json()['response']

if __name__ == '__main__':
    print(get_completion('你好如何应对压力'))

# 模型下载
from modelscope import snapshot_download
model_dir = snapshot_download('qwen/Qwen-1_8B-Chat')

# 可以注释掉选择模型
# llm = LLM(mode='offline').init_model('Linly', 'Linly-AI/Chinese-LLaMA-2-7B-hf')
# llm = LLM(mode='offline').init_model('Gemini', 'gemini-pro', api_key = "your api key")
# llm = LLM(mode='offline').init_model('Qwen', 'Qwen/Qwen-1_8B-Chat')

# 可以通过config来设置模型
llm = LLM(mode=mode).init_model('Qwen', model_path)