diff --git a/.gitattributes b/.gitattributes
index c7d9f3332a950355d5a77d85000f05e6f45435ea..fa0ada747c9d2ecf8c8e9cd4495d4ccf0609f682 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -32,3 +32,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+cleaners/JapaneseCleaner.dll filter=lfs diff=lfs merge=lfs -text
+cleaners/sys.dic filter=lfs diff=lfs merge=lfs -text
diff --git a/README.md b/README.md
index 32897cd3e640101ba184f8c4ccd896981de3804a..5dff1202fc4a2059b8dedea05953f2d01b5fb122 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,13 @@
 ---
+title: Chatbot With Vits
+emoji: 🚀
+colorFrom: purple
+colorTo: indigo
+sdk: gradio
+sdk_version: 3.23.0
+app_file: app.py
+pinned: false
 license: mit
 ---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
diff --git a/app.py b/app.py
new file mode 100644
index 0000000000000000000000000000000000000000..b031ebd991a89850c0b4746e7ad28d2de4dab081
--- /dev/null
+++ b/app.py
@@ -0,0 +1,251 @@
+import logging
+logging.getLogger('numba').setLevel(logging.WARNING)
+logging.getLogger('matplotlib').setLevel(logging.WARNING)
+logging.getLogger('urllib3').setLevel(logging.WARNING)
+from text import text_to_sequence
+import numpy as np
+from scipy.io import wavfile
+import torch
+import json
+import commons
+import utils
+import sys
+import pathlib
+import onnxruntime as ort
+import gradio as gr
+import argparse
+import time
+import os
+import io
+from scipy.io.wavfile import write
+from flask import Flask, request
+from threading import Thread
+import openai
+import requests
+class VitsGradio:
+    def __init__(self):
+        self.lan = ["中文","日文","自动"]
+        self.chatapi = ["gpt-3.5-turbo","gpt3"]
+        self.modelPaths = []
+        for root,dirs,files in os.walk("checkpoints"):
+            for dir in dirs:
+                self.modelPaths.append(dir)
+        with gr.Blocks() as self.Vits:
+            with gr.Tab("调试用"):
+                with gr.Row():
+                    with gr.Column():
+                        with gr.Row():
+                            with gr.Column():
+                                self.text = gr.TextArea(label="Text", value="你好")
+                                with gr.Accordion(label="测试api", open=False):
+                                    self.local_chat1 = gr.Checkbox(value=False, label="使用网址+文本进行模拟")
+                                    self.url_input = gr.TextArea(label="键入测试", value="http://127.0.0.1:8080/chat?Text=")
+                                    butto = gr.Button("测试从网页端获取文本")
+                                btnVC = gr.Button("测试tts+对话程序")
+                            with gr.Column():
+                                output2 = gr.TextArea(label="回复")
+                                output1 = gr.Audio(label="采样率22050")
+                                output3 = gr.outputs.File(label="44100hz: output.wav")
+                butto.click(self.Simul, inputs=[self.text, self.url_input], outputs=[output2,output3])
+                btnVC.click(self.tts_fn, inputs=[self.text], outputs=[output1,output2])
+            with gr.Tab("控制面板"):
+                with gr.Row():
+                    with gr.Column():
+                        with gr.Row():
+                            with gr.Column():
+                                self.api_input1 = gr.TextArea(label="输入api-key或ChATGLM模型的路径", value="https://platform.openai.com/account/api-keys")
+                                with gr.Accordion(label="chatbot选择", open=False):
+                                                self.api_input2 = gr.Checkbox(value=True, label="采用gpt3.5")
+                                                self.local_chat1 = gr.Checkbox(value=False, label="启动本地chatbot")
+                                                self.local_chat2 = gr.Checkbox(value=True, label="是否量化")
+                                                res = gr.TextArea()
+                                Botselection = gr.Button("聊天机器人选择")
+                                Botselection.click(self.check_bot, inputs=[self.api_input1,self.api_input2,self.local_chat1,self.local_chat2], outputs = [res])
+                                self.input1 = gr.Dropdown(label = "vits模型加载", choices = self.modelPaths, value = self.modelPaths[0], type = "value")
+                                self.input2 = gr.Dropdown(label="Language", choices=self.lan, value="自动", interactive=True)
+                            with gr.Column():
+                                btnVC = gr.Button("Submit")
+                                self.input3 = gr.Dropdown(label="Speaker", choices=list(range(101)), value=0, interactive=True)
+                                self.input4 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声比例(noise scale)，以控制情感", value=0.267)
+                                self.input5 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声偏差(noise scale w)，以控制音素长短", value=0.7)
+                                self.input6 = gr.Slider(minimum=0.1, maximum=10, label="duration", value=1)
+                                statusa = gr.TextArea()
+                btnVC.click(self.create_tts_fn, inputs=[self.input1, self.input2, self.input3, self.input4, self.input5, self.input6], outputs = [statusa])
+
+    def Simul(self,text,url_input):
+        web = url_input + text
+        res = requests.get(web)
+        music = res.content
+        with open('output.wav', 'wb') as code:
+            code.write(music)
+        file_path = "output.wav"
+        return web,file_path
+
+
+    def chatgpt(self,text):
+        self.messages.append({"role": "user", "content": text},)
+        chat = openai.ChatCompletion.create(model="gpt-3.5-turbo", messages= self.messages)
+        reply = chat.choices[0].message.content
+        return reply
+    
+    def ChATGLM(self,text):
+        if text == 'clear':
+            self.history = []
+        response, new_history = self.model.chat(self.tokenizer, text, self.history)
+        response = response.replace(" ",'').replace("\n",'.')
+        self.history = new_history
+        return response
+    
+    def gpt3_chat(self,text):
+        call_name = "Waifu"
+        openai.api_key = args.key
+        identity = ""
+        start_sequence = '\n'+str(call_name)+':'
+        restart_sequence = "\nYou: "
+        if 1 == 1:
+            prompt0 = text #当期prompt
+        if text == 'quit':
+            return prompt0
+        prompt = identity + prompt0 + start_sequence
+        response = openai.Completion.create(
+            model="text-davinci-003",
+            prompt=prompt,
+            temperature=0.5,
+            max_tokens=1000,
+            top_p=1.0,
+            frequency_penalty=0.5,
+            presence_penalty=0.0,
+            stop=["\nYou:"]
+        )
+        return response['choices'][0]['text'].strip()
+    
+    def check_bot(self,api_input1,api_input2,local_chat1,local_chat2):
+        if local_chat1:
+            from transformers import AutoTokenizer, AutoModel
+            self.tokenizer = AutoTokenizer.from_pretrained(api_input1, trust_remote_code=True)
+            if local_chat2:
+                self.model = AutoModel.from_pretrained(api_input1, trust_remote_code=True).half().quantize(4).cuda()
+            else:
+                self.model = AutoModel.from_pretrained(api_input1, trust_remote_code=True)
+            self.history = []
+        else:
+            self.messages = []
+            openai.api_key = api_input1
+        return "Finished"
+    
+    def is_japanese(self,string):
+        for ch in string:
+            if ord(ch) > 0x3040 and ord(ch) < 0x30FF:
+                return True
+        return False
+    
+    def is_english(self,string):
+        import re
+        pattern = re.compile('^[A-Za-z0-9.,:;!?()_*"\' ]+$')
+        if pattern.fullmatch(string):
+            return True
+        else:
+            return False
+
+    def get_symbols_from_json(self,path):
+        assert os.path.isfile(path)
+        with open(path, 'r') as f:
+            data = json.load(f)
+        return data['symbols']
+
+    def sle(self,language,text):
+        text = text.replace('\n','。').replace(' ',',')
+        if language == "中文":
+            tts_input1 = "[ZH]" + text + "[ZH]"
+            return tts_input1
+        elif language == "自动":
+            tts_input1 = f"[JA]{text}[JA]" if self.is_japanese(text) else f"[ZH]{text}[ZH]"
+            return tts_input1
+        elif language == "日文":
+            tts_input1 = "[JA]" + text + "[JA]"
+            return tts_input1
+
+    def get_text(self,text,hps_ms):
+        text_norm = text_to_sequence(text,hps_ms.data.text_cleaners)
+        if hps_ms.data.add_blank:
+            text_norm = commons.intersperse(text_norm, 0)
+        text_norm = torch.LongTensor(text_norm)
+        return text_norm
+
+    def create_tts_fn(self,path, input2, input3, n_scale= 0.667,n_scale_w = 0.8, l_scale = 1 ):
+        self.symbols = self.get_symbols_from_json(f"checkpoints/{path}/config.json")
+        self.hps = utils.get_hparams_from_file(f"checkpoints/{path}/config.json")
+        phone_dict = {
+                symbol: i for i, symbol in enumerate(self.symbols)
+            }
+        self.ort_sess = ort.InferenceSession(f"checkpoints/{path}/model.onnx")
+        self.language = input2
+        self.speaker_id = input3
+        self.n_scale = n_scale
+        self.n_scale_w = n_scale_w
+        self.l_scale = l_scale
+        print(self.language,self.speaker_id,self.n_scale)
+        return 'success'
+    
+    def tts_fn(self,text):
+        if self.local_chat1:
+            text = self.chatgpt(text)
+        elif self.api_input2:
+            text = self.ChATGLM(text)
+        else:
+            text = self.gpt3_chat(text)
+        print(text)
+        text =self.sle(self.language,text)
+        seq = text_to_sequence(text, cleaner_names=self.hps.data.text_cleaners)
+        if self.hps.data.add_blank:
+            seq = commons.intersperse(seq, 0)
+        with torch.no_grad():
+            x = np.array([seq], dtype=np.int64)
+            x_len = np.array([x.shape[1]], dtype=np.int64)
+            sid = np.array([self.speaker_id], dtype=np.int64)
+            scales = np.array([self.n_scale, self.n_scale_w, self.l_scale], dtype=np.float32)
+            scales.resize(1, 3)
+            ort_inputs = {
+                        'input': x,
+                        'input_lengths': x_len,
+                        'scales': scales,
+                        'sid': sid
+                    }
+            t1 = time.time()
+            audio = np.squeeze(self.ort_sess.run(None, ort_inputs))
+            audio *= 32767.0 / max(0.01, np.max(np.abs(audio))) * 0.6
+            audio = np.clip(audio, -32767.0, 32767.0)
+            t2 = time.time()
+            spending_time = "推理时间："+str(t2-t1)+"s" 
+            print(spending_time)
+            bytes_wav = bytes()
+            byte_io = io.BytesIO(bytes_wav)
+            wavfile.write('moe/temp1.wav',self.hps.data.sampling_rate, audio.astype(np.int16))
+            cmd = 'ffmpeg -y -i '  + 'moe/temp1.wav' + ' -ar 44100 ' + 'moe/temp2.wav'
+            os.system(cmd)    
+        return (self.hps.data.sampling_rate, audio),text.replace('[JA]','').replace('[ZH]','')
+
+app = Flask(__name__)
+print("开始部署")
+grVits = VitsGradio()
+
+@app.route('/chat')
+def text_api():
+    message = request.args.get('Text','')
+    audio,text = grVits.tts_fn(message)
+    text = text.replace('[JA]','').replace('[ZH]','')
+    with open('moe/temp2.wav','rb') as bit:
+        wav_bytes = bit.read()
+    headers = {
+            'Content-Type': 'audio/wav',
+            'Text': text.encode('utf-8')}
+    return wav_bytes, 200, headers
+
+def gradio_interface():
+    return grVits.Vits.launch()
+
+if __name__ == '__main__':
+    api_thread = Thread(target=app.run, args=("0.0.0.0", 8080))
+    gradio_thread = Thread(target=gradio_interface)
+    api_thread.start()
+    gradio_thread.start()
\ No newline at end of file
diff --git a/attentions.py b/attentions.py
new file mode 100644
index 0000000000000000000000000000000000000000..f8e5112051bae41715bed99a7b6e14ef54b18f60
--- /dev/null
+++ b/attentions.py
@@ -0,0 +1,392 @@
+import math
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+import commons
+from modules import LayerNorm
+
+
+class Encoder(nn.Module):
+    def __init__(self,
+                 hidden_channels,
+                 filter_channels,
+                 n_heads,
+                 n_layers,
+                 kernel_size=1,
+                 p_dropout=0.,
+                 window_size=4,
+                 **kwargs):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+
+        self.drop = nn.Dropout(p_dropout)
+        self.attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.attn_layers.append(
+                MultiHeadAttention(hidden_channels,
+                                   hidden_channels,
+                                   n_heads,
+                                   p_dropout=p_dropout,
+                                   window_size=window_size))
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout))
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask):
+        attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.attn_layers[i](x, x, attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(self,
+                 hidden_channels,
+                 filter_channels,
+                 n_heads,
+                 n_layers,
+                 kernel_size=1,
+                 p_dropout=0.,
+                 proximal_bias=False,
+                 proximal_init=True,
+                 **kwargs):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+
+        self.drop = nn.Dropout(p_dropout)
+        self.self_attn_layers = nn.ModuleList()
+        self.norm_layers_0 = nn.ModuleList()
+        self.encdec_attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.self_attn_layers.append(
+                MultiHeadAttention(hidden_channels,
+                                   hidden_channels,
+                                   n_heads,
+                                   p_dropout=p_dropout,
+                                   proximal_bias=proximal_bias,
+                                   proximal_init=proximal_init))
+            self.norm_layers_0.append(LayerNorm(hidden_channels))
+            self.encdec_attn_layers.append(
+                MultiHeadAttention(hidden_channels,
+                                   hidden_channels,
+                                   n_heads,
+                                   p_dropout=p_dropout))
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                    causal=True))
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, h, h_mask):
+        """
+    x: decoder input
+    h: encoder output
+    """
+        self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
+            device=x.device, dtype=x.dtype)
+        encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.self_attn_layers[i](x, x, self_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_0[i](x + y)
+
+            y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(self,
+                 channels,
+                 out_channels,
+                 n_heads,
+                 p_dropout=0.,
+                 window_size=None,
+                 heads_share=True,
+                 block_length=None,
+                 proximal_bias=False,
+                 proximal_init=False):
+        super().__init__()
+        assert channels % n_heads == 0
+
+        self.channels = channels
+        self.out_channels = out_channels
+        self.n_heads = n_heads
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        self.heads_share = heads_share
+        self.block_length = block_length
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+        self.attn = None
+
+        self.k_channels = channels // n_heads
+        self.conv_q = nn.Conv1d(channels, channels, 1)
+        self.conv_k = nn.Conv1d(channels, channels, 1)
+        self.conv_v = nn.Conv1d(channels, channels, 1)
+        self.conv_o = nn.Conv1d(channels, out_channels, 1)
+        self.drop = nn.Dropout(p_dropout)
+
+        if window_size is not None:
+            n_heads_rel = 1 if heads_share else n_heads
+            rel_stddev = self.k_channels**-0.5
+            self.emb_rel_k = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev)
+            self.emb_rel_v = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev)
+
+        nn.init.xavier_uniform_(self.conv_q.weight)
+        nn.init.xavier_uniform_(self.conv_k.weight)
+        nn.init.xavier_uniform_(self.conv_v.weight)
+        if proximal_init:
+            with torch.no_grad():
+                self.conv_k.weight.copy_(self.conv_q.weight)
+                self.conv_k.bias.copy_(self.conv_q.bias)
+
+    def forward(self, x, c, attn_mask=None):
+        q = self.conv_q(x)
+        k = self.conv_k(c)
+        v = self.conv_v(c)
+
+        x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+        x = self.conv_o(x)
+        return x
+
+    def attention(self, query, key, value, mask=None):
+        # reshape [b, d, t] -> [b, n_h, t, d_k]
+        b, d, t_s, t_t = (*key.size(), query.size(2))
+        query = query.view(b, self.n_heads, self.k_channels,
+                           t_t).transpose(2, 3)
+        key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+        value = value.view(b, self.n_heads, self.k_channels,
+                           t_s).transpose(2, 3)
+
+        scores = torch.matmul(query / math.sqrt(self.k_channels),
+                              key.transpose(-2, -1))
+        if self.window_size is not None:
+            msg = "Relative attention is only available for self-attention."
+            assert t_s == t_t, msg
+            key_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_k, t_s)
+            rel_logits = self._matmul_with_relative_keys(
+                query / math.sqrt(self.k_channels), key_relative_embeddings)
+            scores_local = self._relative_position_to_absolute_position(
+                rel_logits)
+            scores = scores + scores_local
+        if self.proximal_bias:
+            msg = "Proximal bias is only available for self-attention."
+            assert t_s == t_t, msg
+            scores = scores + self._attention_bias_proximal(t_s).to(
+                device=scores.device, dtype=scores.dtype)
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, -1e4)
+            if self.block_length is not None:
+                msg = "Local attention is only available for self-attention."
+                assert t_s == t_t, msg
+                block_mask = torch.ones_like(scores).triu(
+                    -self.block_length).tril(self.block_length)
+                scores = scores.masked_fill(block_mask == 0, -1e4)
+        p_attn = F.softmax(scores, dim=-1)  # [b, n_h, t_t, t_s]
+        p_attn = self.drop(p_attn)
+        output = torch.matmul(p_attn, value)
+        if self.window_size is not None:
+            relative_weights = self._absolute_position_to_relative_position(
+                p_attn)
+            value_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_v, t_s)
+            output = output + self._matmul_with_relative_values(
+                relative_weights, value_relative_embeddings)
+        output = output.transpose(2, 3).contiguous().view(
+            b, d, t_t)  # [b, n_h, t_t, d_k] -> [b, d, t_t]
+        return output, p_attn
+
+    def _matmul_with_relative_values(self, x, y):
+        """
+    x: [b, h, l, m]
+    y: [h or 1, m, d]
+    ret: [b, h, l, d]
+    """
+        ret = torch.matmul(x, y.unsqueeze(0))
+        return ret
+
+    def _matmul_with_relative_keys(self, x, y):
+        """
+    x: [b, h, l, d]
+    y: [h or 1, m, d]
+    ret: [b, h, l, m]
+    """
+        ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+        return ret
+
+    def _get_relative_embeddings(self, relative_embeddings, length):
+        max_relative_position = 2 * self.window_size + 1
+        # Pad first before slice to avoid using cond ops.
+        pad_length = max(length - (self.window_size + 1), 0)
+        slice_start_position = max((self.window_size + 1) - length, 0)
+        slice_end_position = slice_start_position + 2 * length - 1
+        if pad_length > 0:
+            padded_relative_embeddings = F.pad(
+                relative_embeddings,
+                commons.convert_pad_shape([[0, 0], [pad_length, pad_length],
+                                           [0, 0]]))
+        else:
+            padded_relative_embeddings = relative_embeddings
+        used_relative_embeddings = padded_relative_embeddings[:,
+                                                              slice_start_position:
+                                                              slice_end_position]
+        return used_relative_embeddings
+
+    def _relative_position_to_absolute_position(self, x):
+        """
+    x: [b, h, l, 2*l-1]
+    ret: [b, h, l, l]
+    """
+        batch, heads, length, _ = x.size()
+        # Concat columns of pad to shift from relative to absolute indexing.
+        x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0,
+                                                                         1]]))
+
+        # Concat extra elements so to add up to shape (len+1, 2*len-1).
+        x_flat = x.view([batch, heads, length * 2 * length])
+        x_flat = F.pad(
+            x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0,
+                                                                length - 1]]))
+
+        # Reshape and slice out the padded elements.
+        x_final = x_flat.view([batch, heads, length + 1,
+                               2 * length - 1])[:, :, :length, length - 1:]
+        return x_final
+
+    def _absolute_position_to_relative_position(self, x):
+        """
+    x: [b, h, l, l]
+    ret: [b, h, l, 2*l-1]
+    """
+        batch, heads, length, _ = x.size()
+        # padd along column
+        x = F.pad(
+            x,
+            commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0,
+                                                                length - 1]]))
+        x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
+        # add 0's in the beginning that will skew the elements after reshape
+        x_flat = F.pad(
+            x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+        x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
+        return x_final
+
+    def _attention_bias_proximal(self, length):
+        """Bias for self-attention to encourage attention to close positions.
+    Args:
+      length: an integer scalar.
+    Returns:
+      a Tensor with shape [1, 1, length, length]
+    """
+        r = torch.arange(length, dtype=torch.float32)
+        diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+        return torch.unsqueeze(
+            torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 filter_channels,
+                 kernel_size,
+                 p_dropout=0.,
+                 activation=None,
+                 causal=False):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.activation = activation
+        self.causal = causal
+
+        if causal:
+            self.padding = self._causal_padding
+        else:
+            self.padding = self._same_padding
+
+        self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+        self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+        self.drop = nn.Dropout(p_dropout)
+
+    def forward(self, x, x_mask):
+        x = self.conv_1(self.padding(x * x_mask))
+        if self.activation == "gelu":
+            x = x * torch.sigmoid(1.702 * x)
+        else:
+            x = torch.relu(x)
+        x = self.drop(x)
+        x = self.conv_2(self.padding(x * x_mask))
+        return x * x_mask
+
+    def _causal_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = self.kernel_size - 1
+        pad_r = 0
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x
+
+    def _same_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = (self.kernel_size - 1) // 2
+        pad_r = self.kernel_size // 2
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x
diff --git a/checkpoints/Default/config.json b/checkpoints/Default/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1087a33c6ec52fe0eb883c0f352e72af20babc15
--- /dev/null
+++ b/checkpoints/Default/config.json
@@ -0,0 +1,35 @@
+{
+  "train": {
+    "segment_size": 8192
+  },
+  "data": {
+    "text_cleaners":["zh_ja_mixture_cleaners"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "add_blank": true,
+    "n_speakers": 5
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "gin_channels": 256
+  },
+  "speakers": ["\u7dbe\u5730\u5be7\u3005", "\u5728\u539f\u4e03\u6d77", "\u5c0f\u8338", "\u5510\u4e50\u541f"],
+  "symbols": ["_", ",", ".", "!", "?", "-", "~", "\u2026", "A", "E", "I", "N", "O", "Q", "U", "a", "b", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "r", "s", "t", "u", "v", "w", "y", "z", "\u0283", "\u02a7", "\u02a6", "\u026f", "\u0279", "\u0259", "\u0265", "\u207c", "\u02b0", "`", "\u2192", "\u2193", "\u2191", " "]
+}
diff --git a/checkpoints/Default/model.onnx b/checkpoints/Default/model.onnx
new file mode 100644
index 0000000000000000000000000000000000000000..90e50ed41ab7b856959257d0254a3c0bf66e76cd
--- /dev/null
+++ b/checkpoints/Default/model.onnx
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ac210365de160dd5db134f9333525e4ff38426a6a24fcb73e24375b09bef15e
+size 121090654
diff --git a/checkpoints/info.json b/checkpoints/info.json
new file mode 100644
index 0000000000000000000000000000000000000000..96e6c4c5c32c0c259d42b36a9d503aed9fcb3542
--- /dev/null
+++ b/checkpoints/info.json
@@ -0,0 +1,221 @@
+{
+    "Nijigasaki High School":{
+        "speakers":{
+            "高咲侑":{
+                "sid": 0, 
+                "setting": "只选一个做不到啊",
+                "name": "高咲侑"
+            },
+            "歩夢":{
+                "sid": 1,
+                "setting": "只选一个做不到啊",
+                "name": "歩夢"
+            },
+            "かすみ":{
+                "sid": 2,
+                "setting": "只选一个做不到啊",
+                "name": "かすみ"
+            },
+            "しずく":{
+                "sid": 3,
+                "setting": "只选一个做不到啊",
+                "name": "しずく"
+            },
+            "果林":{
+                "sid": 4,
+                "setting": "只选一个做不到啊",
+                "name": "果林"
+            },
+            "愛":{
+                "sid": 5,
+                "setting": "只选一个做不到啊",
+                "name": "愛"
+            },
+            "せつ菜":{
+                "sid": 7,
+                "setting": "只选一个做不到啊",
+                "name": "せつ菜"
+            },
+            "エマ":{
+                "sid": 8,
+                "setting": "只选一个做不到啊",
+                "name": "エマ"
+            },
+            "璃奈":{
+                "sid": 9,
+                "setting": "只选一个做不到啊",
+                "name": "璃奈"
+            },
+            "栞子":{
+                "sid": 10,
+                "setting": "只选一个做不到啊",
+                "name": "栞子"
+            },
+            "ランジュ":{
+                "sid": 11,
+                "setting": "只选一个做不到啊",
+                "name": "ランジュ"
+            },
+            "ミア":{
+                "sid": 12,
+                "setting": "只选一个做不到啊",
+                "name": "ミア"
+            }
+        },
+        "checkpoint":  "checkpoints/Nijigasaki/model.onnx"   
+    },
+    "Seisho Music Academy":{
+        "speakers":{
+            "華恋":{
+                "sid": 21,
+                "setting": "只选一个做不到啊",
+                "name": "華恋"
+            },
+            "まひる":{
+                "sid": 22,
+                "setting": "只选一个做不到啊",
+                "name": "まひる"
+            },
+            "なな":{
+                "sid": 23,
+                "setting": "只选一个做不到啊",
+                "name": "なな"
+            },
+            "クロディーヌ":{
+                "sid": 24,
+                "setting": "只选一个做不到啊",
+                "name": "クロディーヌ"
+            },
+            "ひかり":{
+                "sid": 25,
+                "setting": "只选一个做不到啊",
+                "name": "ひかり"
+            },
+            "純那":{
+                "sid": 26,
+                "setting": "只选一个做不到啊",
+                "name": "純那"
+            },
+            "香子":{
+                "sid": 27,
+                "setting": "只选一个做不到啊",
+                "name": "香子"
+            },
+            "真矢":{
+                "sid": 28,
+                "setting": "只选一个做不到啊",
+                "name": "真矢"
+            },
+            "双葉":{
+                "sid": 29,
+                "setting": "只选一个做不到啊",
+                "name": "双葉"
+            }
+        },
+        "checkpoint": "checkpoints/Starlight/model.onnx"         
+    },
+    "Rinmeikan Girls School":{
+        "speakers":{
+            "珠緒":{
+                "sid": 37,
+                "setting": "只选一个做不到啊",
+                "name": "珠緒"
+            },
+            "塁":{
+                "sid": 36,
+                "setting": "只选一个做不到啊",
+                "name": "塁"
+            },
+            "ゆゆ子":{
+                "sid": 35,
+                "setting": "只选一个做不到啊",
+                "name": "ゆゆ子"
+            },
+            "いちえ":{
+                "sid": 34,
+                "setting": "只选一个做不到啊",
+                "name": "いちえ"
+            }
+        },
+        "checkpoint":  "checkpoints/Starlight/model.onnx"      
+        
+    },
+    "Frontier School of Arts":{
+        "speakers":{
+            "あるる":{
+                "sid": 38,
+                "setting": "只选一个做不到啊",
+                "name": "あるる"
+            },
+            "ララフィン":{
+                "sid": 39,
+                "setting": "只选一个做不到啊",
+                "name": "ララフィン"
+            },
+            "美空":{
+                "sid": 40,
+                "setting": "只选一个做不到啊",
+                "name": "美空"
+            },
+            "静羽":{
+                "sid": 41,
+                "setting": "只选一个做不到啊",
+                "name": "静羽"
+            }
+        },
+        "checkpoint": "checkpoints/Nijigasaki/model.onnx"   
+        
+    },
+    "Siegfeld Institute of Music":{
+        "speakers":{
+            "ミチル":{
+                "sid": 30,
+                "setting": "只选一个做不到啊",
+                "name": "ミチル"
+            },
+            "メイファン":{
+                "sid": 31,
+                "setting": "只选一个做不到啊",
+                "name": "メイファン"
+            },
+            "やちよ":{
+                "sid": 32,
+                "setting": "只选一个做不到啊",
+                "name": "やちよ"
+            },
+            "晶":{
+                "sid": 33,
+                "setting": "只选一个做不到啊",
+                "name": "晶"
+            }
+        },
+        "checkpoint":  "checkpoints/Starlight/model.onnx"    
+        
+    },
+    "Youzusoft":{
+        "speakers":{
+            "宁宁":{
+                "sid": 0,
+                "setting": "只选一个做不到啊",
+                "name": "宁宁"
+            },
+            "在原七海":{
+                "sid": 1,
+                "setting": "只选一个做不到啊",
+                "name": "在原七海"
+            },
+            "小茸":{
+                "sid": 2,
+                "setting": "只选一个做不到啊",
+                "name": "小茸"
+            },
+            "唐乐吟":{
+                "sid": 3,
+                "setting": "只选一个做不到啊",
+                "name": "唐乐吟"
+            }
+        },
+        "checkpoint":  "checkpoints/Default/model.onnx"    
+        
+    }
+}
\ No newline at end of file
diff --git a/cleaners/JapaneseCleaner.dll b/cleaners/JapaneseCleaner.dll
new file mode 100644
index 0000000000000000000000000000000000000000..42ac45fb90957d960232fdfdb4878fd9cf9bc7ba
--- /dev/null
+++ b/cleaners/JapaneseCleaner.dll
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a659eb68d12d4a88ef7dfde6086b9974cd4d43634f7e4bfe710d5537cdd61a75
+size 3097600
diff --git a/cleaners/char.bin b/cleaners/char.bin
new file mode 100644
index 0000000000000000000000000000000000000000..9f6369901c7383d9038c9c49939279e47f6a4db9
--- /dev/null
+++ b/cleaners/char.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:888ee94c5a8a7a26d24ab3f1b7155441351954fd51ea06b4a2f78bd742492b2f
+size 262496
diff --git a/cleaners/matrix.bin b/cleaners/matrix.bin
new file mode 100644
index 0000000000000000000000000000000000000000..bf7a464c5c6768b506986b1f01f8c82bb707b2b8
--- /dev/null
+++ b/cleaners/matrix.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62fd16b4f64c851d5dc352ef0d5740c5fc83ddc7c203b2b0b1fc5271969a14ce
+size 3792262
diff --git a/cleaners/sys.dic b/cleaners/sys.dic
new file mode 100644
index 0000000000000000000000000000000000000000..700526dd4ccbd412d38d057307f5980d1be5a422
--- /dev/null
+++ b/cleaners/sys.dic
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca57d9029691a70a5dfb99afc2844180256161d7130da65b1a867510e129b9a6
+size 103073776
diff --git a/cleaners/unk.dic b/cleaners/unk.dic
new file mode 100644
index 0000000000000000000000000000000000000000..2635af4613f6c6012259743bad97594ab0c85bc0
Binary files /dev/null and b/cleaners/unk.dic differ
diff --git a/commons.py b/commons.py
new file mode 100644
index 0000000000000000000000000000000000000000..33ec83a7986a12b237d28d5e610222881d6b42ae
--- /dev/null
+++ b/commons.py
@@ -0,0 +1,161 @@
+import math
+
+import torch
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+    return int((kernel_size * dilation - dilation) / 2)
+
+
+def convert_pad_shape(pad_shape):
+    pad_shape = [item for sublist in reversed(pad_shape) for item in sublist]
+    return pad_shape
+
+
+def intersperse(lst, item):
+    result = [item] * (len(lst) * 2 + 1)
+    result[1::2] = lst
+    return result
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+    """KL(P||Q)"""
+    kl = (logs_q - logs_p) - 0.5
+    kl += 0.5 * (torch.exp(2. * logs_p) +
+                 ((m_p - m_q)**2)) * torch.exp(-2. * logs_q)
+    return kl
+
+
+def rand_gumbel(shape):
+    """Sample from the Gumbel distribution, protect from overflows."""
+    uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+    return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+    g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+    return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, :, idx_str:idx_end]
+    return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+    b, d, t = x.size()
+    if x_lengths is None:
+        x_lengths = t
+    ids_str_max = x_lengths - segment_size + 1
+    ids_str = (torch.rand([b]).to(device=x.device) *
+               ids_str_max).to(dtype=torch.long)
+    ret = slice_segments(x, ids_str, segment_size)
+    return ret, ids_str
+
+
+def get_timing_signal_1d(length,
+                         channels,
+                         min_timescale=1.0,
+                         max_timescale=1.0e4):
+    position = torch.arange(length, dtype=torch.float)
+    num_timescales = channels // 2
+    log_timescale_increment = (
+        math.log(float(max_timescale) / float(min_timescale)) /
+        (num_timescales - 1))
+    inv_timescales = min_timescale * torch.exp(
+        torch.arange(num_timescales, dtype=torch.float) *
+        -log_timescale_increment)
+    scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+    signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+    signal = F.pad(signal, [0, 0, 0, channels % 2])
+    signal = signal.view(1, channels, length)
+    return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale,
+                                  max_timescale)
+    return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale,
+                                  max_timescale)
+    return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+    mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+    return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+def shift_1d(x):
+    x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+    return x
+
+
+def sequence_mask(length, max_length=None):
+    if max_length is None:
+        max_length = length.max()
+    x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+    return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+    """
+  duration: [b, 1, t_x]
+  mask: [b, 1, t_y, t_x]
+  """
+    device = duration.device
+
+    b, _, t_y, t_x = mask.shape
+    cum_duration = torch.cumsum(duration, -1)
+
+    cum_duration_flat = cum_duration.view(b * t_x)
+    path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+    path = path.view(b, t_x, t_y)
+    path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]
+                                                 ]))[:, :-1]
+    path = path.unsqueeze(1).transpose(2, 3) * mask
+    return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+    if isinstance(parameters, torch.Tensor):
+        parameters = [parameters]
+    parameters = list(filter(lambda p: p.grad is not None, parameters))
+    norm_type = float(norm_type)
+    if clip_value is not None:
+        clip_value = float(clip_value)
+
+    total_norm = 0
+    for p in parameters:
+        param_norm = p.grad.data.norm(norm_type)
+        total_norm += param_norm.item()**norm_type
+        if clip_value is not None:
+            p.grad.data.clamp_(min=-clip_value, max=clip_value)
+    total_norm = total_norm**(1. / norm_type)
+    return total_norm
diff --git a/data_utils.py b/data_utils.py
new file mode 100644
index 0000000000000000000000000000000000000000..1d10e95398af4e76ae61de15fca8e0e80d85282d
--- /dev/null
+++ b/data_utils.py
@@ -0,0 +1,307 @@
+import os
+import random
+
+import torch
+import torchaudio
+import torch.utils.data
+
+import commons
+from mel_processing import spectrogram_torch
+from utils import load_filepaths_and_text
+
+
+class TextAudioSpeakerLoader(torch.utils.data.Dataset):
+    """
+        1) loads audio, speaker_id, text pairs
+        2) normalizes text and converts them to sequences of integers
+        3) computes spectrograms from audio files.
+    """
+    def __init__(self, audiopaths_sid_text, hparams):
+        self.audiopaths_sid_text = load_filepaths_and_text(audiopaths_sid_text)
+        # self.text_cleaners = hparams.text_cleaners
+        self.max_wav_value = hparams.max_wav_value
+        self.sampling_rate = hparams.sampling_rate
+        self.filter_length = hparams.filter_length
+        self.hop_length = hparams.hop_length
+        self.win_length = hparams.win_length
+        self.sampling_rate = hparams.sampling_rate
+        self.src_sampling_rate = getattr(hparams, "src_sampling_rate",
+                                         self.sampling_rate)
+
+        self.cleaned_text = getattr(hparams, "cleaned_text", False)
+
+        self.add_blank = hparams.add_blank
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 190)
+
+        phone_file = getattr(hparams, "phone_table", None)
+        self.phone_dict = None
+        if phone_file is not None:
+            self.phone_dict = {}
+            with open(phone_file) as fin:
+                for line in fin:
+                    arr = line.strip().split()
+                    self.phone_dict[arr[0]] = int(arr[1])
+
+        speaker_file = getattr(hparams, "speaker_table", None)
+        self.speaker_dict = None
+        if speaker_file is not None:
+            self.speaker_dict = {}
+            with open(speaker_file) as fin:
+                for line in fin:
+                    arr = line.strip().split()
+                    self.speaker_dict[arr[0]] = int(arr[1])
+
+        random.seed(1234)
+        random.shuffle(self.audiopaths_sid_text)
+        self._filter()
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+
+        audiopaths_sid_text_new = []
+        lengths = []
+        for item in self.audiopaths_sid_text:
+            audiopath = item[0]
+            # filename|text or filename|speaker|text
+            text = item[1] if len(item) == 2 else item[2]
+            if self.min_text_len <= len(text) and len(
+                    text) <= self.max_text_len:
+                audiopaths_sid_text_new.append(item)
+                lengths.append(
+                    int(
+                        os.path.getsize(audiopath) * self.sampling_rate /
+                        self.src_sampling_rate) // (2 * self.hop_length))
+        self.audiopaths_sid_text = audiopaths_sid_text_new
+        self.lengths = lengths
+
+    def get_audio_text_speaker_pair(self, audiopath_sid_text):
+        audiopath = audiopath_sid_text[0]
+        if len(audiopath_sid_text) == 2:  # filename|text
+            sid = 0
+            text = audiopath_sid_text[1]
+        else:  # filename|speaker|text
+            sid = self.speaker_dict[audiopath_sid_text[1]]
+            text = audiopath_sid_text[2]
+        text = self.get_text(text)
+        spec, wav = self.get_audio(audiopath)
+        sid = self.get_sid(sid)
+        return (text, spec, wav, sid)
+
+    def get_audio(self, filename):
+        audio, sampling_rate = torchaudio.load(filename, normalize=False)
+        if sampling_rate != self.sampling_rate:
+            audio = audio.to(torch.float)
+            audio = torchaudio.transforms.Resample(sampling_rate,
+                                                   self.sampling_rate)(audio)
+            audio = audio.to(torch.int16)
+        audio = audio[0]  # Get the first channel
+        audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec = spectrogram_torch(audio_norm,
+                                 self.filter_length,
+                                 self.sampling_rate,
+                                 self.hop_length,
+                                 self.win_length,
+                                 center=False)
+        spec = torch.squeeze(spec, 0)
+        return spec, audio_norm
+
+    def get_text(self, text):
+        text_norm = [self.phone_dict[phone] for phone in text.split()]
+        if self.add_blank:
+            text_norm = commons.intersperse(text_norm, 0)
+        text_norm = torch.LongTensor(text_norm)
+        return text_norm
+
+    def get_sid(self, sid):
+        sid = torch.LongTensor([int(sid)])
+        return sid
+
+    def __getitem__(self, index):
+        return self.get_audio_text_speaker_pair(
+            self.audiopaths_sid_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_sid_text)
+
+
+class TextAudioSpeakerCollate():
+    """ Zero-pads model inputs and targets
+    """
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text, audio and speaker identities
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized, sid]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(torch.LongTensor(
+            [x[1].size(1) for x in batch]),
+            dim=0,
+            descending=True)
+
+        max_text_len = max([len(x[0]) for x in batch])
+        max_spec_len = max([x[1].size(1) for x in batch])
+        max_wav_len = max([x[2].size(1) for x in batch])
+
+        text_lengths = torch.LongTensor(len(batch))
+        spec_lengths = torch.LongTensor(len(batch))
+        wav_lengths = torch.LongTensor(len(batch))
+        sid = torch.LongTensor(len(batch))
+
+        text_padded = torch.LongTensor(len(batch), max_text_len)
+        spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0),
+                                        max_spec_len)
+        wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
+        text_padded.zero_()
+        spec_padded.zero_()
+        wav_padded.zero_()
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            text = row[0]
+            text_padded[i, :text.size(0)] = text
+            text_lengths[i] = text.size(0)
+
+            spec = row[1]
+            spec_padded[i, :, :spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wav = row[2]
+            wav_padded[i, :, :wav.size(1)] = wav
+            wav_lengths[i] = wav.size(1)
+
+            sid[i] = row[3]
+
+        if self.return_ids:
+            return (text_padded, text_lengths, spec_padded, spec_lengths,
+                    wav_padded, wav_lengths, sid, ids_sorted_decreasing)
+        return (text_padded, text_lengths, spec_padded, spec_lengths,
+                wav_padded, wav_lengths, sid)
+
+
+class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler
+                               ):
+    """
+    Maintain similar input lengths in a batch.
+    Length groups are specified by boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any batch is included either
+    {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}.
+
+    It removes samples which are not included in the boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1
+    or length(x) > b3 are discarded.
+    """
+    def __init__(self,
+                 dataset,
+                 batch_size,
+                 boundaries,
+                 num_replicas=None,
+                 rank=None,
+                 shuffle=True):
+        super().__init__(dataset,
+                         num_replicas=num_replicas,
+                         rank=rank,
+                         shuffle=shuffle)
+        self.lengths = dataset.lengths
+        self.batch_size = batch_size
+        self.boundaries = boundaries
+
+        self.buckets, self.num_samples_per_bucket = self._create_buckets()
+        self.total_size = sum(self.num_samples_per_bucket)
+        self.num_samples = self.total_size // self.num_replicas
+
+    def _create_buckets(self):
+        buckets = [[] for _ in range(len(self.boundaries) - 1)]
+        for i in range(len(self.lengths)):
+            length = self.lengths[i]
+            idx_bucket = self._bisect(length)
+            if idx_bucket != -1:
+                buckets[idx_bucket].append(i)
+
+        for i in range(len(buckets) - 1, 0, -1):
+            if len(buckets[i]) == 0:
+                buckets.pop(i)
+                self.boundaries.pop(i + 1)
+
+        num_samples_per_bucket = []
+        for i in range(len(buckets)):
+            len_bucket = len(buckets[i])
+            total_batch_size = self.num_replicas * self.batch_size
+            rem = (total_batch_size -
+                   (len_bucket % total_batch_size)) % total_batch_size
+            num_samples_per_bucket.append(len_bucket + rem)
+        return buckets, num_samples_per_bucket
+
+    def __iter__(self):
+        # deterministically shuffle based on epoch
+        g = torch.Generator()
+        g.manual_seed(self.epoch)
+
+        indices = []
+        if self.shuffle:
+            for bucket in self.buckets:
+                indices.append(
+                    torch.randperm(len(bucket), generator=g).tolist())
+        else:
+            for bucket in self.buckets:
+                indices.append(list(range(len(bucket))))
+
+        batches = []
+        for i in range(len(self.buckets)):
+            bucket = self.buckets[i]
+            len_bucket = len(bucket)
+            ids_bucket = indices[i]
+            num_samples_bucket = self.num_samples_per_bucket[i]
+
+            # add extra samples to make it evenly divisible
+            rem = num_samples_bucket - len_bucket
+            ids_bucket = ids_bucket + ids_bucket * (
+                rem // len_bucket) + ids_bucket[:(rem % len_bucket)]
+
+            # subsample
+            ids_bucket = ids_bucket[self.rank::self.num_replicas]
+
+            # batching
+            for j in range(len(ids_bucket) // self.batch_size):
+                batch = [
+                    bucket[idx]
+                    for idx in ids_bucket[j * self.batch_size:(j + 1) *
+                                          self.batch_size]
+                ]
+                batches.append(batch)
+
+        if self.shuffle:
+            batch_ids = torch.randperm(len(batches), generator=g).tolist()
+            batches = [batches[i] for i in batch_ids]
+        self.batches = batches
+
+        assert len(self.batches) * self.batch_size == self.num_samples
+        return iter(self.batches)
+
+    def _bisect(self, x, lo=0, hi=None):
+        if hi is None:
+            hi = len(self.boundaries) - 1
+
+        if hi > lo:
+            mid = (hi + lo) // 2
+            if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]:
+                return mid
+            elif x <= self.boundaries[mid]:
+                return self._bisect(x, lo, mid)
+            else:
+                return self._bisect(x, mid + 1, hi)
+        else:
+            return -1
+
+    def __len__(self):
+        return self.num_samples // self.batch_size
diff --git a/export_onnx.py b/export_onnx.py
new file mode 100644
index 0000000000000000000000000000000000000000..1834d24bc581ee43af038c2f56668d7a8bb1e425
--- /dev/null
+++ b/export_onnx.py
@@ -0,0 +1,140 @@
+# Copyright (c) 2022, Yongqiang Li (yongqiangli@alumni.hust.edu.cn)
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import json
+import os
+import sys
+
+import torch
+
+from models import SynthesizerTrn
+import utils
+
+try:
+    import onnxruntime as ort
+except ImportError:
+    print('Please install onnxruntime!')
+    sys.exit(1)
+
+
+def to_numpy(tensor):
+    return tensor.detach().cpu().numpy() if tensor.requires_grad \
+        else tensor.detach().numpy()
+
+
+def get_args():
+    parser = argparse.ArgumentParser(description='export onnx model')
+    parser.add_argument('--checkpoint', required=True, help='checkpoint')
+    parser.add_argument('--cfg', required=True, help='config file')
+    parser.add_argument('--onnx_model', required=True, help='onnx model name')
+    # parser.add_argument('--phone_table',
+    #                     required=True,
+    #                     help='input phone dict')
+    # parser.add_argument('--speaker_table', default=None, help='speaker table')
+    # parser.add_argument("--speaker_num", required=True,
+    #                     type=int, help="speaker num")
+    parser.add_argument(
+        '--providers',
+        required=False,
+        default='CPUExecutionProvider',
+        choices=['CUDAExecutionProvider', 'CPUExecutionProvider'],
+        help='the model to send request to')
+    args = parser.parse_args()
+    return args
+
+
+def get_data_from_cfg(cfg_path: str):
+    assert os.path.isfile(cfg_path)
+    with open(cfg_path, 'r') as f:
+        data = json.load(f)
+        symbols = data["symbols"]
+        speaker_num = data["data"]["n_speakers"]
+    return len(symbols), speaker_num
+
+
+def main():
+    args = get_args()
+    os.environ['CUDA_VISIBLE_DEVICES'] = '0'
+
+    hps = utils.get_hparams_from_file(args.cfg)
+    # with open(args.phone_table) as p_f:
+    #     phone_num = len(p_f.readlines()) + 1
+    # num_speakers = 1
+    # if args.speaker_table is not None:
+    #     num_speakers = len(open(args.speaker_table).readlines()) + 1
+    phone_num, num_speakers = get_data_from_cfg(args.cfg)
+    net_g = SynthesizerTrn(phone_num,
+                           hps.data.filter_length // 2 + 1,
+                           hps.train.segment_size // hps.data.hop_length,
+                           n_speakers=num_speakers,
+                           **hps.model)
+    utils.load_checkpoint(args.checkpoint, net_g, None)
+    net_g.forward = net_g.export_forward
+    net_g.eval()
+
+    seq = torch.randint(low=0, high=phone_num, size=(1, 10), dtype=torch.long)
+    seq_len = torch.IntTensor([seq.size(1)]).long()
+
+    # noise(可用于控制感情等变化程度) lenth(可用于控制整体语速) noisew(控制音素发音长度变化程度)
+    # 参考 https://github.com/gbxh/genshinTTS
+    scales = torch.FloatTensor([0.667, 1.0, 0.8])
+    # make triton dynamic shape happy
+    scales = scales.unsqueeze(0)
+    sid = torch.IntTensor([0]).long()
+
+    dummy_input = (seq, seq_len, scales, sid)
+    torch.onnx.export(model=net_g,
+                      args=dummy_input,
+                      f=args.onnx_model,
+                      input_names=['input', 'input_lengths', 'scales', 'sid'],
+                      output_names=['output'],
+                      dynamic_axes={
+                          'input': {
+                              0: 'batch',
+                              1: 'phonemes'
+                          },
+                          'input_lengths': {
+                              0: 'batch'
+                          },
+                          'scales': {
+                              0: 'batch'
+                          },
+                          'sid': {
+                              0: 'batch'
+                          },
+                          'output': {
+                              0: 'batch',
+                              1: 'audio',
+                              2: 'audio_length'
+                          }
+                      },
+                      opset_version=13,
+                      verbose=False)
+
+    # Verify onnx precision
+    torch_output = net_g(seq, seq_len, scales, sid)
+    providers = [args.providers]
+    ort_sess = ort.InferenceSession(args.onnx_model, providers=providers)
+    ort_inputs = {
+        'input': to_numpy(seq),
+        'input_lengths': to_numpy(seq_len),
+        'scales': to_numpy(scales),
+        'sid': to_numpy(sid),
+    }
+    onnx_output = ort_sess.run(None, ort_inputs)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/losses.py b/losses.py
new file mode 100644
index 0000000000000000000000000000000000000000..f835539a16b49e1065fef4e4a1efb259b88dcf64
--- /dev/null
+++ b/losses.py
@@ -0,0 +1,58 @@
+import torch
+
+
+def feature_loss(fmap_r, fmap_g):
+    loss = 0
+    for dr, dg in zip(fmap_r, fmap_g):
+        for rl, gl in zip(dr, dg):
+            rl = rl.float().detach()
+            gl = gl.float()
+            loss += torch.mean(torch.abs(rl - gl))
+
+    return loss * 2
+
+
+def discriminator_loss(disc_real_outputs, disc_generated_outputs):
+    loss = 0
+    r_losses = []
+    g_losses = []
+    for dr, dg in zip(disc_real_outputs, disc_generated_outputs):
+        dr = dr.float()
+        dg = dg.float()
+        r_loss = torch.mean((1 - dr)**2)
+        g_loss = torch.mean(dg**2)
+        loss += (r_loss + g_loss)
+        r_losses.append(r_loss.item())
+        g_losses.append(g_loss.item())
+
+    return loss, r_losses, g_losses
+
+
+def generator_loss(disc_outputs):
+    loss = 0
+    gen_losses = []
+    for dg in disc_outputs:
+        dg = dg.float()
+        l = torch.mean((1 - dg)**2)
+        gen_losses.append(l)
+        loss += l
+
+    return loss, gen_losses
+
+
+def kl_loss(z_p, logs_q, m_p, logs_p, z_mask):
+    """
+  z_p, logs_q: [b, h, t_t]
+  m_p, logs_p: [b, h, t_t]
+  """
+    z_p = z_p.float()
+    logs_q = logs_q.float()
+    m_p = m_p.float()
+    logs_p = logs_p.float()
+    z_mask = z_mask.float()
+
+    kl = logs_p - logs_q - 0.5
+    kl += 0.5 * ((z_p - m_p)**2) * torch.exp(-2. * logs_p)
+    kl = torch.sum(kl * z_mask)
+    l = kl / torch.sum(z_mask)
+    return l
diff --git a/main.py b/main.py
new file mode 100644
index 0000000000000000000000000000000000000000..54d633252ac3087ff8bfa4753af5d4997acd8e5f
--- /dev/null
+++ b/main.py
@@ -0,0 +1,255 @@
+import logging
+logging.getLogger('numba').setLevel(logging.WARNING)
+logging.getLogger('matplotlib').setLevel(logging.WARNING)
+logging.getLogger('urllib3').setLevel(logging.WARNING)
+opj = input("是否启用pyopenjtalk?封装版无法保证非japanese cleaners推理日语时的质量(Y/N)")
+if opj == "N":
+    from TEXTS import text_to_sequence
+else:
+    from text import text_to_sequence
+import numpy as np
+from scipy.io import wavfile
+import torch
+import json
+import commons
+import utils
+import sys
+import pathlib
+import onnxruntime as ort
+import gradio as gr
+import argparse
+import time
+import os
+import io
+from scipy.io.wavfile import write
+from flask import Flask, request
+from threading import Thread
+import openai
+import requests
+class VitsGradio:
+    def __init__(self):
+        self.lan = ["中文","日文","自动"]
+        self.chatapi = ["gpt-3.5-turbo","gpt3"]
+        self.modelPaths = []
+        for root,dirs,files in os.walk("checkpoints"):
+            for dir in dirs:
+                self.modelPaths.append(dir)
+        with gr.Blocks() as self.Vits:
+            with gr.Tab("调试用"):
+                with gr.Row():
+                    with gr.Column():
+                        with gr.Row():
+                            with gr.Column():
+                                self.text = gr.TextArea(label="Text", value="你好")
+                                with gr.Accordion(label="测试api", open=False):
+                                    self.local_chat1 = gr.Checkbox(value=False, label="使用网址+文本进行模拟")
+                                    self.url_input = gr.TextArea(label="键入测试", value="http://127.0.0.1:8080/chat?Text=")
+                                    butto = gr.Button("测试从网页端获取文本")
+                                btnVC = gr.Button("测试tts+对话程序")
+                            with gr.Column():
+                                output2 = gr.TextArea(label="回复")
+                                output1 = gr.Audio(label="采样率22050")
+                                output3 = gr.outputs.File(label="44100hz: output.wav")
+                butto.click(self.Simul, inputs=[self.text, self.url_input], outputs=[output2,output3])
+                btnVC.click(self.tts_fn, inputs=[self.text], outputs=[output1,output2])
+            with gr.Tab("控制面板"):
+                with gr.Row():
+                    with gr.Column():
+                        with gr.Row():
+                            with gr.Column():
+                                self.api_input1 = gr.TextArea(label="输入api-key或本地存储说话模型的路径", value="https://platform.openai.com/account/api-keys")
+                                with gr.Accordion(label="chatbot选择", open=False):
+                                                self.api_input2 = gr.Checkbox(value=True, label="采用gpt3.5")
+                                                self.local_chat1 = gr.Checkbox(value=False, label="启动本地chatbot")
+                                                self.local_chat2 = gr.Checkbox(value=True, label="是否量化")
+                                                res = gr.TextArea()
+                                Botselection = gr.Button("确认模型")
+                                Botselection.click(self.check_bot, inputs=[self.api_input1,self.api_input2,self.local_chat1,self.local_chat2], outputs = [res])
+                                self.input1 = gr.Dropdown(label = "模型", choices = self.modelPaths, value = self.modelPaths[0], type = "value")
+                                self.input2 = gr.Dropdown(label="Language", choices=self.lan, value="自动", interactive=True)
+                            with gr.Column():
+                                btnVC = gr.Button("Submit")
+                                self.input3 = gr.Dropdown(label="Speaker", choices=list(range(101)), value=0, interactive=True)
+                                self.input4 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声比例(noise scale)，以控制情感", value=0.267)
+                                self.input5 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声偏差(noise scale w)，以控制音素长短", value=0.7)
+                                self.input6 = gr.Slider(minimum=0.1, maximum=10, label="duration", value=1)
+                                statusa = gr.TextArea()
+                btnVC.click(self.create_tts_fn, inputs=[self.input1, self.input2, self.input3, self.input4, self.input5, self.input6], outputs = [statusa])
+
+    def Simul(self,text,url_input):
+        web = url_input + text
+        res = requests.get(web)
+        music = res.content
+        with open('output.wav', 'wb') as code:
+            code.write(music)
+        file_path = "output.wav"
+        return web,file_path
+
+
+    def chatgpt(self,text):
+        self.messages.append({"role": "user", "content": text},)
+        chat = openai.ChatCompletion.create(model="gpt-3.5-turbo", messages= self.messages)
+        reply = chat.choices[0].message.content
+        return reply
+    
+    def ChATGLM(self,text):
+        if text == 'clear':
+            self.history = []
+        response, new_history = self.model.chat(self.tokenizer, text, self.history)
+        response = response.replace(" ",'').replace("\n",'.')
+        self.history = new_history
+        return response
+    
+    def gpt3_chat(self,text):
+        call_name = "Waifu"
+        openai.api_key = args.key
+        identity = ""
+        start_sequence = '\n'+str(call_name)+':'
+        restart_sequence = "\nYou: "
+        if 1 == 1:
+            prompt0 = text #当期prompt
+        if text == 'quit':
+            return prompt0
+        prompt = identity + prompt0 + start_sequence
+        response = openai.Completion.create(
+            model="text-davinci-003",
+            prompt=prompt,
+            temperature=0.5,
+            max_tokens=1000,
+            top_p=1.0,
+            frequency_penalty=0.5,
+            presence_penalty=0.0,
+            stop=["\nYou:"]
+        )
+        return response['choices'][0]['text'].strip()
+    
+    def check_bot(self,api_input1,api_input2,local_chat1,local_chat2):
+        if local_chat1:
+            from transformers import AutoTokenizer, AutoModel
+            self.tokenizer = AutoTokenizer.from_pretrained(api_input1, trust_remote_code=True)
+            if local_chat2:
+                self.model = AutoModel.from_pretrained(api_input1, trust_remote_code=True).half().quantize(4).cuda()
+            else:
+                self.model = AutoModel.from_pretrained(api_input1, trust_remote_code=True)
+            self.history = []
+        else:
+            self.messages = []
+            openai.api_key = api_input1
+        return "Finished"
+    
+    def is_japanese(self,string):
+        for ch in string:
+            if ord(ch) > 0x3040 and ord(ch) < 0x30FF:
+                return True
+        return False
+    
+    def is_english(self,string):
+        import re
+        pattern = re.compile('^[A-Za-z0-9.,:;!?()_*"\' ]+$')
+        if pattern.fullmatch(string):
+            return True
+        else:
+            return False
+
+    def get_symbols_from_json(self,path):
+        assert os.path.isfile(path)
+        with open(path, 'r') as f:
+            data = json.load(f)
+        return data['symbols']
+
+    def sle(self,language,text):
+        text = text.replace('\n','。').replace(' ',',')
+        if language == "中文":
+            tts_input1 = "[ZH]" + text + "[ZH]"
+            return tts_input1
+        elif language == "自动":
+            tts_input1 = f"[JA]{text}[JA]" if self.is_japanese(text) else f"[ZH]{text}[ZH]"
+            return tts_input1
+        elif language == "日文":
+            tts_input1 = "[JA]" + text + "[JA]"
+            return tts_input1
+
+    def get_text(self,text,hps_ms):
+        text_norm = text_to_sequence(text,hps_ms.data.text_cleaners)
+        if hps_ms.data.add_blank:
+            text_norm = commons.intersperse(text_norm, 0)
+        text_norm = torch.LongTensor(text_norm)
+        return text_norm
+
+    def create_tts_fn(self,path, input2, input3, n_scale= 0.667,n_scale_w = 0.8, l_scale = 1 ):
+        self.symbols = self.get_symbols_from_json(f"checkpoints/{path}/config.json")
+        self.hps = utils.get_hparams_from_file(f"checkpoints/{path}/config.json")
+        phone_dict = {
+                symbol: i for i, symbol in enumerate(self.symbols)
+            }
+        self.ort_sess = ort.InferenceSession(f"checkpoints/{path}/model.onnx")
+        self.language = input2
+        self.speaker_id = input3
+        self.n_scale = n_scale
+        self.n_scale_w = n_scale_w
+        self.l_scale = l_scale
+        print(self.language,self.speaker_id,self.n_scale)
+        return 'success'
+    
+    def tts_fn(self,text):
+        if self.local_chat1:
+            text = self.chatgpt(text)
+        elif self.api_input2:
+            text = self.ChATGLM(text)
+        else:
+            text = self.gpt3_chat(text)
+        print(text)
+        text =self.sle(self.language,text)
+        seq = text_to_sequence(text, cleaner_names=self.hps.data.text_cleaners)
+        if self.hps.data.add_blank:
+            seq = commons.intersperse(seq, 0)
+        with torch.no_grad():
+            x = np.array([seq], dtype=np.int64)
+            x_len = np.array([x.shape[1]], dtype=np.int64)
+            sid = np.array([self.speaker_id], dtype=np.int64)
+            scales = np.array([self.n_scale, self.n_scale_w, self.l_scale], dtype=np.float32)
+            scales.resize(1, 3)
+            ort_inputs = {
+                        'input': x,
+                        'input_lengths': x_len,
+                        'scales': scales,
+                        'sid': sid
+                    }
+            t1 = time.time()
+            audio = np.squeeze(self.ort_sess.run(None, ort_inputs))
+            audio *= 32767.0 / max(0.01, np.max(np.abs(audio))) * 0.6
+            audio = np.clip(audio, -32767.0, 32767.0)
+            t2 = time.time()
+            spending_time = "推理时间："+str(t2-t1)+"s" 
+            print(spending_time)
+            bytes_wav = bytes()
+            byte_io = io.BytesIO(bytes_wav)
+            wavfile.write('moe/temp1.wav',self.hps.data.sampling_rate, audio.astype(np.int16))
+            cmd = 'ffmpeg -y -i '  + 'moe/temp1.wav' + ' -ar 44100 ' + 'moe/temp2.wav'
+            os.system(cmd)    
+        return (self.hps.data.sampling_rate, audio),text.replace('[JA]','').replace('[ZH]','')
+
+app = Flask(__name__)
+print("开始部署")
+grVits = VitsGradio()
+
+@app.route('/chat')
+def text_api():
+    message = request.args.get('Text','')
+    audio,text = grVits.tts_fn(message)
+    text = text.replace('[JA]','').replace('[ZH]','')
+    with open('moe/temp2.wav','rb') as bit:
+        wav_bytes = bit.read()
+    headers = {
+            'Content-Type': 'audio/wav',
+            'Text': text.encode('utf-8')}
+    return wav_bytes, 200, headers
+
+def gradio_interface():
+    return grVits.Vits.launch()
+
+if __name__ == '__main__':
+    api_thread = Thread(target=app.run, args=("0.0.0.0", 8080))
+    gradio_thread = Thread(target=gradio_interface)
+    api_thread.start()
+    gradio_thread.start()
\ No newline at end of file
diff --git a/mel_processing.py b/mel_processing.py
new file mode 100644
index 0000000000000000000000000000000000000000..575ee5ef554ce4ef12cfee0cfbae87c869c1be05
--- /dev/null
+++ b/mel_processing.py
@@ -0,0 +1,137 @@
+import torch
+import torch.nn.functional as F
+import torch.utils.data
+from librosa.filters import mel as librosa_mel_fn
+
+MAX_WAV_VALUE = 32768.0
+
+
+def dynamic_range_compression_torch(x, C=1, clip_val=1e-5):
+    """
+    PARAMS
+    ------
+    C: compression factor
+    """
+    return torch.log(torch.clamp(x, min=clip_val) * C)
+
+
+def dynamic_range_decompression_torch(x, C=1):
+    """
+    PARAMS
+    ------
+    C: compression factor used to compress
+    """
+    return torch.exp(x) / C
+
+
+def spectral_normalize_torch(magnitudes):
+    output = dynamic_range_compression_torch(magnitudes)
+    return output
+
+
+def spectral_de_normalize_torch(magnitudes):
+    output = dynamic_range_decompression_torch(magnitudes)
+    return output
+
+
+mel_basis = {}
+hann_window = {}
+
+
+def spectrogram_torch(y,
+                      n_fft,
+                      sampling_rate,
+                      hop_size,
+                      win_size,
+                      center=False):
+    if torch.min(y) < -1.:
+        print('min value is ', torch.min(y))
+    if torch.max(y) > 1.:
+        print('max value is ', torch.max(y))
+
+    global hann_window
+    dtype_device = str(y.dtype) + '_' + str(y.device)
+    wnsize_dtype_device = str(win_size) + '_' + dtype_device
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(
+            dtype=y.dtype, device=y.device)
+
+    y = F.pad(y.unsqueeze(1),
+              (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)),
+              mode='reflect')
+    y = y.squeeze(1)
+
+    spec = torch.stft(y,
+                      n_fft,
+                      hop_length=hop_size,
+                      win_length=win_size,
+                      window=hann_window[wnsize_dtype_device],
+                      center=center,
+                      pad_mode='reflect',
+                      normalized=False,
+                      onesided=True)
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+    return spec
+
+
+def spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax):
+    global mel_basis
+    dtype_device = str(spec.dtype) + '_' + str(spec.device)
+    fmax_dtype_device = str(fmax) + '_' + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sampling_rate, n_fft, num_mels, fmin, fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(
+            dtype=spec.dtype, device=spec.device)
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+    return spec
+
+
+def mel_spectrogram_torch(y,
+                          n_fft,
+                          num_mels,
+                          sampling_rate,
+                          hop_size,
+                          win_size,
+                          fmin,
+                          fmax,
+                          center=False):
+    if torch.min(y) < -1.:
+        print('min value is ', torch.min(y))
+    if torch.max(y) > 1.:
+        print('max value is ', torch.max(y))
+
+    global mel_basis, hann_window
+    dtype_device = str(y.dtype) + '_' + str(y.device)
+    fmax_dtype_device = str(fmax) + '_' + dtype_device
+    wnsize_dtype_device = str(win_size) + '_' + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sampling_rate, n_fft, num_mels, fmin, fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(
+            dtype=y.dtype, device=y.device)
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(
+            dtype=y.dtype, device=y.device)
+
+    y = F.pad(y.unsqueeze(1),
+              (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)),
+              mode='reflect')
+    y = y.squeeze(1)
+
+    spec = torch.stft(y,
+                      n_fft,
+                      hop_length=hop_size,
+                      win_length=win_size,
+                      window=hann_window[wnsize_dtype_device],
+                      center=center,
+                      pad_mode='reflect',
+                      normalized=False,
+                      onesided=True)
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+
+    return spec
diff --git a/models.py b/models.py
new file mode 100644
index 0000000000000000000000000000000000000000..7757ad7afecddf87b29e120ba3784e9bed42d713
--- /dev/null
+++ b/models.py
@@ -0,0 +1,672 @@
+import math
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+from torch.nn import Conv1d, ConvTranspose1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+import monotonic_align
+
+import commons
+import modules
+import attentions
+from commons import init_weights, get_padding
+
+
+class StochasticDurationPredictor(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 filter_channels,
+                 kernel_size,
+                 p_dropout,
+                 n_flows=4,
+                 gin_channels=0):
+        super().__init__()
+        filter_channels = in_channels  # it needs to be removed from future version.
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.log_flow = modules.Log()
+        self.flows = nn.ModuleList()
+        self.flows.append(modules.ElementwiseAffine(2))
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3))
+            self.flows.append(modules.Flip())
+
+        self.post_pre = nn.Conv1d(1, filter_channels, 1)
+        self.post_proj = nn.Conv1d(filter_channels, filter_channels, 1)
+        self.post_convs = modules.DDSConv(filter_channels,
+                                          kernel_size,
+                                          n_layers=3,
+                                          p_dropout=p_dropout)
+        self.post_flows = nn.ModuleList()
+        self.post_flows.append(modules.ElementwiseAffine(2))
+        for i in range(4):
+            self.post_flows.append(
+                modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3))
+            self.post_flows.append(modules.Flip())
+
+        self.pre = nn.Conv1d(in_channels, filter_channels, 1)
+        self.proj = nn.Conv1d(filter_channels, filter_channels, 1)
+        self.convs = modules.DDSConv(filter_channels,
+                                     kernel_size,
+                                     n_layers=3,
+                                     p_dropout=p_dropout)
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, filter_channels, 1)
+
+    def forward(self,
+                x,
+                x_mask,
+                w=None,
+                g=None,
+                reverse=False,
+                noise_scale=1.0):
+        x = torch.detach(x)
+        x = self.pre(x)
+        if g is not None:
+            g = torch.detach(g)
+            x = x + self.cond(g)
+        x = self.convs(x, x_mask)
+        x = self.proj(x) * x_mask
+
+        if not reverse:
+            flows = self.flows
+            assert w is not None
+
+            logdet_tot_q = 0
+            h_w = self.post_pre(w)
+            h_w = self.post_convs(h_w, x_mask)
+            h_w = self.post_proj(h_w) * x_mask
+            e_q = torch.randn(w.size(0), 2, w.size(2)).to(
+                device=x.device, dtype=x.dtype) * x_mask
+            z_q = e_q
+            for flow in self.post_flows:
+                z_q, logdet_q = flow(z_q, x_mask, g=(x + h_w))
+                logdet_tot_q += logdet_q
+            z_u, z1 = torch.split(z_q, [1, 1], 1)
+            u = torch.sigmoid(z_u) * x_mask
+            z0 = (w - u) * x_mask
+            logdet_tot_q += torch.sum(
+                (F.logsigmoid(z_u) + F.logsigmoid(-z_u)) * x_mask, [1, 2])
+            logq = torch.sum(
+                -0.5 * (math.log(2 * math.pi) +
+                        (e_q**2)) * x_mask, [1, 2]) - logdet_tot_q
+
+            logdet_tot = 0
+            z0, logdet = self.log_flow(z0, x_mask)
+            logdet_tot += logdet
+            z = torch.cat([z0, z1], 1)
+            for flow in flows:
+                z, logdet = flow(z, x_mask, g=x, reverse=reverse)
+                logdet_tot = logdet_tot + logdet
+            nll = torch.sum(0.5 * (math.log(2 * math.pi) +
+                                   (z**2)) * x_mask, [1, 2]) - logdet_tot
+            return nll + logq  # [b]
+        else:
+            flows = list(reversed(self.flows))
+            flows = flows[:-2] + [flows[-1]]  # remove a useless vflow
+            z = torch.randn(x.size(0), 2, x.size(2)).to(
+                device=x.device, dtype=x.dtype) * noise_scale
+            for flow in flows:
+                z = flow(z, x_mask, g=x, reverse=reverse)
+            z0, z1 = torch.split(z, [1, 1], 1)
+            logw = z0
+            return logw
+
+
+class DurationPredictor(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 filter_channels,
+                 kernel_size,
+                 p_dropout,
+                 gin_channels=0):
+        super().__init__()
+
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.gin_channels = gin_channels
+
+        self.drop = nn.Dropout(p_dropout)
+        self.conv_1 = nn.Conv1d(in_channels,
+                                filter_channels,
+                                kernel_size,
+                                padding=kernel_size // 2)
+        self.norm_1 = modules.LayerNorm(filter_channels)
+        self.conv_2 = nn.Conv1d(filter_channels,
+                                filter_channels,
+                                kernel_size,
+                                padding=kernel_size // 2)
+        self.norm_2 = modules.LayerNorm(filter_channels)
+        self.proj = nn.Conv1d(filter_channels, 1, 1)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, in_channels, 1)
+
+    def forward(self, x, x_mask, g=None):
+        x = torch.detach(x)
+        if g is not None:
+            g = torch.detach(g)
+            x = x + self.cond(g)
+        x = self.conv_1(x * x_mask)
+        x = torch.relu(x)
+        x = self.norm_1(x)
+        x = self.drop(x)
+        x = self.conv_2(x * x_mask)
+        x = torch.relu(x)
+        x = self.norm_2(x)
+        x = self.drop(x)
+        x = self.proj(x * x_mask)
+        return x * x_mask
+
+
+class TextEncoder(nn.Module):
+    def __init__(self, n_vocab, out_channels, hidden_channels, filter_channels,
+                 n_heads, n_layers, kernel_size, p_dropout):
+        super().__init__()
+        self.n_vocab = n_vocab
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+
+        self.emb = nn.Embedding(n_vocab, hidden_channels)
+        nn.init.normal_(self.emb.weight, 0.0, hidden_channels**-0.5)
+
+        self.encoder = attentions.Encoder(hidden_channels, filter_channels,
+                                          n_heads, n_layers, kernel_size,
+                                          p_dropout)
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths):
+        x = self.emb(x) * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)),
+                                 1).to(x.dtype)
+
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return x, m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(self,
+                 channels,
+                 hidden_channels,
+                 kernel_size,
+                 dilation_rate,
+                 n_layers,
+                 n_flows=4,
+                 gin_channels=0):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ResidualCouplingLayer(channels,
+                                              hidden_channels,
+                                              kernel_size,
+                                              dilation_rate,
+                                              n_layers,
+                                              gin_channels=gin_channels,
+                                              mean_only=True))
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+
+class PosteriorEncoder(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 hidden_channels,
+                 kernel_size,
+                 dilation_rate,
+                 n_layers,
+                 gin_channels=0):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(hidden_channels,
+                              kernel_size,
+                              dilation_rate,
+                              n_layers,
+                              gin_channels=gin_channels)
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, g=None):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)),
+                                 1).to(x.dtype)
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+
+class Generator(torch.nn.Module):
+    def __init__(self,
+                 initial_channel,
+                 resblock,
+                 resblock_kernel_sizes,
+                 resblock_dilation_sizes,
+                 upsample_rates,
+                 upsample_initial_channel,
+                 upsample_kernel_sizes,
+                 gin_channels=0):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(initial_channel,
+                               upsample_initial_channel,
+                               7,
+                               1,
+                               padding=3)
+        resblock = modules.ResBlock1 if resblock == '1' else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(upsample_initial_channel // (2**i),
+                                    upsample_initial_channel // (2**(i + 1)),
+                                    k,
+                                    u,
+                                    padding=(k - u) // 2)))
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2**(i + 1))
+            for j, (k, d) in enumerate(
+                    zip(resblock_kernel_sizes, resblock_dilation_sizes)):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        print('Removing weight norm...')
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self,
+                 period,
+                 kernel_size=5,
+                 stride=3,
+                 use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm is False else spectral_norm
+        self.convs = nn.ModuleList([
+            norm_f(
+                Conv2d(1,
+                       32, (kernel_size, 1), (stride, 1),
+                       padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(
+                Conv2d(32,
+                       128, (kernel_size, 1), (stride, 1),
+                       padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(
+                Conv2d(128,
+                       512, (kernel_size, 1), (stride, 1),
+                       padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(
+                Conv2d(512,
+                       1024, (kernel_size, 1), (stride, 1),
+                       padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(
+                Conv2d(1024,
+                       1024, (kernel_size, 1),
+                       1,
+                       padding=(get_padding(kernel_size, 1), 0))),
+        ])
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm is False else spectral_norm
+        self.convs = nn.ModuleList([
+            norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+            norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+            norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+            norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+            norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+            norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+        ])
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2, 3, 5, 7, 11]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm)
+            for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class SynthesizerTrn(nn.Module):
+    """
+  Synthesizer for Training
+  """
+    def __init__(self,
+                 n_vocab,
+                 spec_channels,
+                 segment_size,
+                 inter_channels,
+                 hidden_channels,
+                 filter_channels,
+                 n_heads,
+                 n_layers,
+                 kernel_size,
+                 p_dropout,
+                 resblock,
+                 resblock_kernel_sizes,
+                 resblock_dilation_sizes,
+                 upsample_rates,
+                 upsample_initial_channel,
+                 upsample_kernel_sizes,
+                 n_speakers=0,
+                 gin_channels=0,
+                 use_sdp=True,
+                 **kwargs):
+
+        super().__init__()
+        self.n_vocab = n_vocab
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.n_speakers = n_speakers
+        self.gin_channels = gin_channels
+        if self.n_speakers != 0:
+            message = "gin_channels must be none zero for multiple speakers"
+            assert gin_channels != 0, message
+
+        self.use_sdp = use_sdp
+
+        self.enc_p = TextEncoder(n_vocab, inter_channels, hidden_channels,
+                                 filter_channels, n_heads, n_layers,
+                                 kernel_size, p_dropout)
+        self.dec = Generator(inter_channels,
+                             resblock,
+                             resblock_kernel_sizes,
+                             resblock_dilation_sizes,
+                             upsample_rates,
+                             upsample_initial_channel,
+                             upsample_kernel_sizes,
+                             gin_channels=gin_channels)
+        self.enc_q = PosteriorEncoder(spec_channels,
+                                      inter_channels,
+                                      hidden_channels,
+                                      5,
+                                      1,
+                                      16,
+                                      gin_channels=gin_channels)
+        self.flow = ResidualCouplingBlock(inter_channels,
+                                          hidden_channels,
+                                          5,
+                                          1,
+                                          4,
+                                          gin_channels=gin_channels)
+
+        if use_sdp:
+            self.dp = StochasticDurationPredictor(hidden_channels,
+                                                  192,
+                                                  3,
+                                                  0.5,
+                                                  4,
+                                                  gin_channels=gin_channels)
+        else:
+            self.dp = DurationPredictor(hidden_channels,
+                                        256,
+                                        3,
+                                        0.5,
+                                        gin_channels=gin_channels)
+
+        if n_speakers > 1:
+            self.emb_g = nn.Embedding(n_speakers, gin_channels)
+
+    def forward(self, x, x_lengths, y, y_lengths, sid=None):
+
+        x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths)
+        if self.n_speakers > 0:
+            g = self.emb_g(sid).unsqueeze(-1)  # [b, h, 1]
+        else:
+            g = None
+
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+
+        with torch.no_grad():
+            # negative cross-entropy
+            s_p_sq_r = torch.exp(-2 * logs_p)  # [b, d, t]
+            neg_cent1 = torch.sum(-0.5 * math.log(2 * math.pi) - logs_p, [1],
+                                  keepdim=True)  # [b, 1, t_s]
+            neg_cent2 = torch.matmul(
+                -0.5 * (z_p**2).transpose(1, 2),
+                s_p_sq_r)  # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s]
+            neg_cent3 = torch.matmul(
+                z_p.transpose(1, 2),
+                (m_p * s_p_sq_r))  # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s]
+            neg_cent4 = torch.sum(-0.5 * (m_p**2) * s_p_sq_r, [1],
+                                  keepdim=True)  # [b, 1, t_s]
+            neg_cent = neg_cent1 + neg_cent2 + neg_cent3 + neg_cent4
+
+            attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(
+                y_mask, -1)
+            attn = monotonic_align.maximum_path(
+                neg_cent, attn_mask.squeeze(1)).unsqueeze(1).detach()
+
+        w = attn.sum(2)
+        if self.use_sdp:
+            l_length = self.dp(x, x_mask, w, g=g)
+            l_length = l_length / torch.sum(x_mask)
+        else:
+            logw_ = torch.log(w + 1e-6) * x_mask
+            logw = self.dp(x, x_mask, g=g)
+            l_length = torch.sum(
+                (logw - logw_)**2, [1, 2]) / torch.sum(x_mask)  # for averaging
+
+        # expand prior
+        m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1,
+                                                          2)).transpose(1, 2)
+        logs_p = torch.matmul(attn.squeeze(1),
+                              logs_p.transpose(1, 2)).transpose(1, 2)
+
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size)
+        o = self.dec(z_slice, g=g)
+        return o, l_length, attn, ids_slice, x_mask, y_mask, (z, z_p, m_p,
+                                                              logs_p, m_q,
+                                                              logs_q)
+
+    def infer(self,
+              x,
+              x_lengths,
+              sid=None,
+              noise_scale=1,
+              length_scale=1,
+              noise_scale_w=1.,
+              max_len=None):
+        x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths)
+        if self.n_speakers > 0:
+            g = self.emb_g(sid).unsqueeze(-1)  # [b, h, 1]
+        else:
+            g = None
+
+        if self.use_sdp:
+            logw = self.dp(x,
+                           x_mask,
+                           g=g,
+                           reverse=True,
+                           noise_scale=noise_scale_w)
+        else:
+            logw = self.dp(x, x_mask, g=g)
+        w = torch.exp(logw) * x_mask * length_scale
+        w_ceil = torch.ceil(w)
+        y_lengths = torch.clamp_min(torch.sum(w_ceil, [1, 2]), 1).long()
+        y_mask = torch.unsqueeze(commons.sequence_mask(y_lengths, None),
+                                 1).to(x_mask.dtype)
+        attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
+        attn = commons.generate_path(w_ceil, attn_mask)
+
+        m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(
+            1, 2)  # [b, t', t], [b, t, d] -> [b, d, t']
+        logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(
+            1, 2)).transpose(1, 2)  # [b, t', t], [b, t, d] -> [b, d, t']
+
+        z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * noise_scale
+        z = self.flow(z_p, y_mask, g=g, reverse=True)
+        o = self.dec((z * y_mask)[:, :, :max_len], g=g)
+        return o, attn, y_mask, (z, z_p, m_p, logs_p)
+
+    def export_forward(self, x, x_lengths, scales, sid):
+        # shape of scales: Bx3, make triton happy
+        audio, *_ = self.infer(x,
+                               x_lengths,
+                               sid,
+                               noise_scale=scales[0][0],
+                               length_scale=scales[0][1],
+                               noise_scale_w=scales[0][2])
+        return audio
+
+    def voice_conversion(self, y, y_lengths, sid_src, sid_tgt):
+        assert self.n_speakers > 0, "n_speakers have to be larger than 0."
+        g_src = self.emb_g(sid_src).unsqueeze(-1)
+        g_tgt = self.emb_g(sid_tgt).unsqueeze(-1)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g_src)
+        z_p = self.flow(z, y_mask, g=g_src)
+        z_hat = self.flow(z_p, y_mask, g=g_tgt, reverse=True)
+        o_hat = self.dec(z_hat * y_mask, g=g_tgt)
+        return o_hat, y_mask, (z, z_p, z_hat)
diff --git a/modules.py b/modules.py
new file mode 100644
index 0000000000000000000000000000000000000000..0f9286cd02b123d54f38e5c2bb144cc01056d080
--- /dev/null
+++ b/modules.py
@@ -0,0 +1,469 @@
+import math
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+from torch.nn import Conv1d
+from torch.nn.utils import weight_norm, remove_weight_norm
+
+import commons
+from commons import init_weights, get_padding
+from transforms import piecewise_rational_quadratic_transform
+
+LRELU_SLOPE = 0.1
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels, ), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+class ConvReluNorm(nn.Module):
+    def __init__(self, in_channels, hidden_channels, out_channels, kernel_size,
+                 n_layers, p_dropout):
+        super().__init__()
+        self.in_channels = in_channels
+        self.hidden_channels = hidden_channels
+        self.out_channels = out_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+        assert n_layers > 1, "Number of layers should be larger than 0."
+
+        self.conv_layers = nn.ModuleList()
+        self.norm_layers = nn.ModuleList()
+        self.conv_layers.append(
+            nn.Conv1d(in_channels,
+                      hidden_channels,
+                      kernel_size,
+                      padding=kernel_size // 2))
+        self.norm_layers.append(LayerNorm(hidden_channels))
+        self.relu_drop = nn.Sequential(nn.ReLU(), nn.Dropout(p_dropout))
+        for _ in range(n_layers - 1):
+            self.conv_layers.append(
+                nn.Conv1d(hidden_channels,
+                          hidden_channels,
+                          kernel_size,
+                          padding=kernel_size // 2))
+            self.norm_layers.append(LayerNorm(hidden_channels))
+        self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask):
+        x_org = x
+        for i in range(self.n_layers):
+            x = self.conv_layers[i](x * x_mask)
+            x = self.norm_layers[i](x)
+            x = self.relu_drop(x)
+        x = x_org + self.proj(x)
+        return x * x_mask
+
+
+class DDSConv(nn.Module):
+    """
+  Dialted and Depth-Separable Convolution
+  """
+    def __init__(self, channels, kernel_size, n_layers, p_dropout=0.):
+        super().__init__()
+        self.channels = channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+
+        self.drop = nn.Dropout(p_dropout)
+        self.convs_sep = nn.ModuleList()
+        self.convs_1x1 = nn.ModuleList()
+        self.norms_1 = nn.ModuleList()
+        self.norms_2 = nn.ModuleList()
+        for i in range(n_layers):
+            dilation = kernel_size**i
+            padding = (kernel_size * dilation - dilation) // 2
+            self.convs_sep.append(
+                nn.Conv1d(channels,
+                          channels,
+                          kernel_size,
+                          groups=channels,
+                          dilation=dilation,
+                          padding=padding))
+            self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
+            self.norms_1.append(LayerNorm(channels))
+            self.norms_2.append(LayerNorm(channels))
+
+    def forward(self, x, x_mask, g=None):
+        if g is not None:
+            x = x + g
+        for i in range(self.n_layers):
+            y = self.convs_sep[i](x * x_mask)
+            y = self.norms_1[i](y)
+            y = F.gelu(y)
+            y = self.convs_1x1[i](y)
+            y = self.norms_2[i](y)
+            y = F.gelu(y)
+            y = self.drop(y)
+            x = x + y
+        return x * x_mask
+
+
+class WN(torch.nn.Module):
+    def __init__(self,
+                 hidden_channels,
+                 kernel_size,
+                 dilation_rate,
+                 n_layers,
+                 gin_channels=0,
+                 p_dropout=0):
+        super(WN, self).__init__()
+        assert (kernel_size % 2 == 1)
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size,
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+        self.p_dropout = p_dropout
+
+        self.in_layers = torch.nn.ModuleList()
+        self.res_skip_layers = torch.nn.ModuleList()
+        self.drop = nn.Dropout(p_dropout)
+
+        if gin_channels != 0:
+            cond_layer = torch.nn.Conv1d(gin_channels,
+                                         2 * hidden_channels * n_layers, 1)
+            self.cond_layer = torch.nn.utils.weight_norm(cond_layer,
+                                                         name='weight')
+
+        for i in range(n_layers):
+            dilation = dilation_rate**i
+            padding = int((kernel_size * dilation - dilation) / 2)
+            in_layer = torch.nn.Conv1d(hidden_channels,
+                                       2 * hidden_channels,
+                                       kernel_size,
+                                       dilation=dilation,
+                                       padding=padding)
+            in_layer = torch.nn.utils.weight_norm(in_layer, name='weight')
+            self.in_layers.append(in_layer)
+
+            # last one is not necessary
+            if i < n_layers - 1:
+                res_skip_channels = 2 * hidden_channels
+            else:
+                res_skip_channels = hidden_channels
+
+            res_skip_layer = torch.nn.Conv1d(hidden_channels,
+                                             res_skip_channels, 1)
+            res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer,
+                                                        name='weight')
+            self.res_skip_layers.append(res_skip_layer)
+
+    def forward(self, x, x_mask, g=None, **kwargs):
+        output = torch.zeros_like(x)
+        n_channels_tensor = torch.IntTensor([self.hidden_channels])
+
+        if g is not None:
+            g = self.cond_layer(g)
+
+        for i in range(self.n_layers):
+            x_in = self.in_layers[i](x)
+            if g is not None:
+                cond_offset = i * 2 * self.hidden_channels
+                g_l = g[:,
+                        cond_offset:cond_offset + 2 * self.hidden_channels, :]
+            else:
+                g_l = torch.zeros_like(x_in)
+
+            acts = commons.fused_add_tanh_sigmoid_multiply(
+                x_in, g_l, n_channels_tensor)
+            acts = self.drop(acts)
+
+            res_skip_acts = self.res_skip_layers[i](acts)
+            if i < self.n_layers - 1:
+                res_acts = res_skip_acts[:, :self.hidden_channels, :]
+                x = (x + res_acts) * x_mask
+                output = output + res_skip_acts[:, self.hidden_channels:, :]
+            else:
+                output = output + res_skip_acts
+        return output * x_mask
+
+    def remove_weight_norm(self):
+        if self.gin_channels != 0:
+            torch.nn.utils.remove_weight_norm(self.cond_layer)
+        for l in self.in_layers:
+            torch.nn.utils.remove_weight_norm(l)
+        for l in self.res_skip_layers:
+            torch.nn.utils.remove_weight_norm(l)
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.convs1 = nn.ModuleList([
+            weight_norm(
+                Conv1d(channels,
+                       channels,
+                       kernel_size,
+                       1,
+                       dilation=dilation[0],
+                       padding=get_padding(kernel_size, dilation[0]))),
+            weight_norm(
+                Conv1d(channels,
+                       channels,
+                       kernel_size,
+                       1,
+                       dilation=dilation[1],
+                       padding=get_padding(kernel_size, dilation[1]))),
+            weight_norm(
+                Conv1d(channels,
+                       channels,
+                       kernel_size,
+                       1,
+                       dilation=dilation[2],
+                       padding=get_padding(kernel_size, dilation[2])))
+        ])
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList([
+            weight_norm(
+                Conv1d(channels,
+                       channels,
+                       kernel_size,
+                       1,
+                       dilation=1,
+                       padding=get_padding(kernel_size, 1))),
+            weight_norm(
+                Conv1d(channels,
+                       channels,
+                       kernel_size,
+                       1,
+                       dilation=1,
+                       padding=get_padding(kernel_size, 1))),
+            weight_norm(
+                Conv1d(channels,
+                       channels,
+                       kernel_size,
+                       1,
+                       dilation=1,
+                       padding=get_padding(kernel_size, 1)))
+        ])
+        self.convs2.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c2(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.convs = nn.ModuleList([
+            weight_norm(
+                Conv1d(channels,
+                       channels,
+                       kernel_size,
+                       1,
+                       dilation=dilation[0],
+                       padding=get_padding(kernel_size, dilation[0]))),
+            weight_norm(
+                Conv1d(channels,
+                       channels,
+                       kernel_size,
+                       1,
+                       dilation=dilation[1],
+                       padding=get_padding(kernel_size, dilation[1])))
+        ])
+        self.convs.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c in self.convs:
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+
+class Log(nn.Module):
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
+            logdet = torch.sum(-y, [1, 2])
+            return y, logdet
+        else:
+            x = torch.exp(x) * x_mask
+            return x
+
+
+class Flip(nn.Module):
+    def forward(self, x, *args, reverse=False, **kwargs):
+        x = torch.flip(x, [1])
+        if not reverse:
+            logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
+            return x, logdet
+        else:
+            return x
+
+
+class ElementwiseAffine(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.channels = channels
+        self.m = nn.Parameter(torch.zeros(channels, 1))
+        self.logs = nn.Parameter(torch.zeros(channels, 1))
+
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = self.m + torch.exp(self.logs) * x
+            y = y * x_mask
+            logdet = torch.sum(self.logs * x_mask, [1, 2])
+            return y, logdet
+        else:
+            x = (x - self.m) * torch.exp(-self.logs) * x_mask
+            return x
+
+
+class ResidualCouplingLayer(nn.Module):
+    def __init__(self,
+                 channels,
+                 hidden_channels,
+                 kernel_size,
+                 dilation_rate,
+                 n_layers,
+                 p_dropout=0,
+                 gin_channels=0,
+                 mean_only=False):
+        assert channels % 2 == 0, "channels should be divisible by 2"
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.half_channels = channels // 2
+        self.mean_only = mean_only
+
+        self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+        self.enc = WN(hidden_channels,
+                      kernel_size,
+                      dilation_rate,
+                      n_layers,
+                      p_dropout=p_dropout,
+                      gin_channels=gin_channels)
+        self.post = nn.Conv1d(hidden_channels,
+                              self.half_channels * (2 - mean_only), 1)
+        self.post.weight.data.zero_()
+        self.post.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0) * x_mask
+        h = self.enc(h, x_mask, g=g)
+        stats = self.post(h) * x_mask
+        if not self.mean_only:
+            m, logs = torch.split(stats, [self.half_channels] * 2, 1)
+        else:
+            m = stats
+            logs = torch.zeros_like(m)
+
+        if not reverse:
+            x1 = m + x1 * torch.exp(logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            logdet = torch.sum(logs, [1, 2])
+            return x, logdet
+        else:
+            x1 = (x1 - m) * torch.exp(-logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            return x
+
+
+class ConvFlow(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 filter_channels,
+                 kernel_size,
+                 n_layers,
+                 num_bins=10,
+                 tail_bound=5.0):
+        super().__init__()
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.num_bins = num_bins
+        self.tail_bound = tail_bound
+        self.half_channels = in_channels // 2
+
+        self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
+        self.convs = DDSConv(filter_channels,
+                             kernel_size,
+                             n_layers,
+                             p_dropout=0.)
+        self.proj = nn.Conv1d(filter_channels,
+                              self.half_channels * (num_bins * 3 - 1), 1)
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0)
+        h = self.convs(h, x_mask, g=g)
+        h = self.proj(h) * x_mask
+
+        b, c, t = x0.shape
+        h = h.reshape(b, c, -1, t).permute(0, 1, 3,
+                                           2)  # [b, cx?, t] -> [b, c, t, ?]
+
+        unnormalized_widths = h[..., :self.num_bins] / math.sqrt(
+            self.filter_channels)
+        unnormalized_heights = h[...,
+                                 self.num_bins:2 * self.num_bins] / math.sqrt(
+                                     self.filter_channels)
+        unnormalized_derivatives = h[..., 2 * self.num_bins:]
+
+        x1, logabsdet = piecewise_rational_quadratic_transform(
+            x1,
+            unnormalized_widths,
+            unnormalized_heights,
+            unnormalized_derivatives,
+            inverse=reverse,
+            tails='linear',
+            tail_bound=self.tail_bound)
+
+        x = torch.cat([x0, x1], 1) * x_mask
+        logdet = torch.sum(logabsdet * x_mask, [1, 2])
+        if not reverse:
+            return x, logdet
+        else:
+            return x
diff --git a/moe/config.json b/moe/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..a907a4d270fc3456b076018f28b57b9925e56750
--- /dev/null
+++ b/moe/config.json
@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "THUDM/chatglm-6b",
+  "architectures": [
+    "ChatGLMModel"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_chatglm.ChatGLMConfig",
+    "AutoModel": "modeling_chatglm.ChatGLMForConditionalGeneration",
+    "AutoModelForSeq2SeqLM": "modeling_chatglm.ChatGLMForConditionalGeneration"
+  },
+  "bos_token_id": 150004,
+  "eos_token_id": 150005,
+  "hidden_size": 4096,
+  "inner_hidden_size": 16384,
+  "layernorm_epsilon": 1e-05,
+  "max_sequence_length": 2048,
+  "model_type": "chatglm",
+  "num_attention_heads": 32,
+  "num_layers": 28,
+  "position_encoding_2d": true,
+  "torch_dtype": "float16",
+  "transformers_version": "4.23.1",
+  "use_cache": true,
+  "vocab_size": 150528
+}
diff --git a/moe/configuration_chatglm.py b/moe/configuration_chatglm.py
new file mode 100644
index 0000000000000000000000000000000000000000..b4b196a51c922f44936b6f6bd2bbd80980f1755d
--- /dev/null
+++ b/moe/configuration_chatglm.py
@@ -0,0 +1,92 @@
+""" ChatGLM model configuration """
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class ChatGLMConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`~ChatGLMModel`].
+    It is used to instantiate an ChatGLM model according to the specified arguments, defining the model
+    architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of
+    the ChatGLM-6B [THUDM/ChatGLM-6B](https://huggingface.co/THUDM/chatglm-6b) architecture.
+
+    Configuration objects inherit from  [`PretrainedConfig`] and can be used
+    to control the model outputs. Read the documentation from  [`PretrainedConfig`]
+    for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 150528):
+            Vocabulary size of the ChatGLM-6B model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`~ChatGLMModel`] or
+            [`~TFChatGLMModel`].
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the encoder layers and the pooler layer.
+        num_hidden_layers (`int`, *optional*, defaults to 28):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        inner_hidden_size (`int`, *optional*, defaults to 16384):
+            Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        max_sequence_length (`int`, *optional*, defaults to 512):
+            The maximum sequence length that this model might ever be used with.
+            Typically set this to something large just in case (e.g., 512 or 1024 or 2048).
+        layernorm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether the model should return the last key/values attentions (not used by all models).
+        Example:
+
+    ```python
+    >>> from configuration_chatglm import ChatGLMConfig
+    >>> from modeling_chatglm import ChatGLMModel
+
+    >>> # Initializing a ChatGLM-6B THUDM/ChatGLM-6B style configuration
+    >>> configuration = ChatGLMConfig()
+
+    >>> # Initializing a model from the THUDM/ChatGLM-6B style configuration
+    >>> model = ChatGLMModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```
+"""
+    model_type = "chatglm"
+
+    def __init__(
+            self,
+            vocab_size=150528,
+            hidden_size=4096,
+            num_layers=28,
+            num_attention_heads=32,
+            layernorm_epsilon=1e-5,
+            use_cache=False,
+            bos_token_id=150004,
+            eos_token_id=150005,
+            pad_token_id=0,
+            max_sequence_length=2048,
+            inner_hidden_size=16384,
+            position_encoding_2d=True,
+            **kwargs
+    ):
+        self.num_layers = num_layers
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_attention_heads = num_attention_heads
+        self.max_sequence_length = max_sequence_length
+        self.layernorm_epsilon = layernorm_epsilon
+        self.inner_hidden_size = inner_hidden_size
+        self.use_cache = use_cache
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.position_encoding_2d = position_encoding_2d
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            **kwargs
+        )
diff --git a/moe/modeling_chatglm.py b/moe/modeling_chatglm.py
new file mode 100644
index 0000000000000000000000000000000000000000..1429dbb820ce1d625e69c471a772f868b867e529
--- /dev/null
+++ b/moe/modeling_chatglm.py
@@ -0,0 +1,1157 @@
+""" PyTorch ChatGLM model. """
+
+import math
+import copy
+import os
+
+import torch
+import torch.utils.checkpoint
+import torch.nn.functional as F
+from torch import nn
+from torch.nn import CrossEntropyLoss, LayerNorm
+from torch.nn.utils import skip_init
+from typing import Optional, Tuple, Union, List
+
+from transformers.utils import (
+    add_code_sample_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+    BaseModelOutputWithPastAndCrossAttentions,
+)
+from transformers.modeling_utils import PreTrainedModel
+
+from transformers.utils import logging
+from .configuration_chatglm import ChatGLMConfig
+
+# flags required to enable jit fusion kernels
+torch._C._jit_set_profiling_mode(False)
+torch._C._jit_set_profiling_executor(False)
+torch._C._jit_override_can_fuse_on_cpu(True)
+torch._C._jit_override_can_fuse_on_gpu(True)
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "THUDM/ChatGLM-6B"
+_CONFIG_FOR_DOC = "ChatGLM6BConfig"
+
+CHATGLM_6B_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "THUDM/chatglm-6b",
+    # See all ChatGLM-6B models at https://huggingface.co/models?filter=chatglm
+]
+
+
+def load_tf_weights_in_chatglm_6b(model, config, tf_checkpoint_path):
+    """Load tf checkpoints in a pytorch model."""
+    try:
+        import re
+
+        import numpy as np
+        import tensorflow as tf
+    except ImportError:
+        logger.error(
+            "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
+            "https://www.tensorflow.org/install/ for installation instructions."
+        )
+        raise
+    tf_path = os.path.abspath(tf_checkpoint_path)
+    logger.info(f"Converting TensorFlow checkpoint from {tf_path}")
+    # Load weights from TF model
+    init_vars = tf.train.list_variables(tf_path)
+    names = []
+    arrays = []
+    for name, shape in init_vars:
+        logger.info(f"Loading TF weight {name} with shape {shape}")
+        array = tf.train.load_variable(tf_path, name)
+        names.append(name)
+        arrays.append(array)
+
+    for name, array in zip(names, arrays):
+        name = name.split("/")
+        # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
+        # which are not required for using pretrained model
+        if any(
+                n in ["adam_v", "adam_m", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "global_step"]
+                for n in name
+        ):
+            logger.info(f"Skipping {'/'.join(name)}")
+            continue
+        pointer = model
+        for m_name in name:
+            if re.fullmatch(r"[A-Za-z]+_\d+", m_name):
+                scope_names = re.split(r"_(\d+)", m_name)
+            else:
+                scope_names = [m_name]
+            if scope_names[0] == "kernel" or scope_names[0] == "gamma":
+                pointer = getattr(pointer, "weight")
+            elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
+                pointer = getattr(pointer, "bias")
+            elif scope_names[0] == "output_weights":
+                pointer = getattr(pointer, "weight")
+            elif scope_names[0] == "squad":
+                pointer = getattr(pointer, "classifier")
+            else:
+                try:
+                    pointer = getattr(pointer, scope_names[0])
+                except AttributeError:
+                    logger.info(f"Skipping {'/'.join(name)}")
+                    continue
+            if len(scope_names) >= 2:
+                num = int(scope_names[1])
+                pointer = pointer[num]
+        if m_name[-11:] == "_embeddings":
+            pointer = getattr(pointer, "weight")
+        elif m_name == "kernel":
+            array = np.transpose(array)
+        try:
+            assert (
+                    pointer.shape == array.shape
+            ), f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched"
+        except AssertionError as e:
+            e.args += (pointer.shape, array.shape)
+            raise
+        logger.info(f"Initialize PyTorch weight {name}")
+        pointer.data = torch.from_numpy(array)
+    return model
+
+
+@torch.jit.script
+def gelu_impl(x):
+    """OpenAI's gelu implementation."""
+    return 0.5 * x * (1.0 + torch.tanh(0.7978845608028654 * x *
+                                       (1.0 + 0.044715 * x * x)))
+
+
+def gelu(x):
+    return gelu_impl(x)
+
+
+class RotaryEmbedding(torch.nn.Module):
+    def __init__(self, dim, base=10000, precision=torch.half, learnable=False):
+        super().__init__()
+        inv_freq = 1. / (base ** (torch.arange(0, dim, 2).float() / dim))
+        inv_freq = inv_freq.half()
+        self.learnable = learnable
+        if learnable:
+            self.inv_freq = torch.nn.Parameter(inv_freq)
+            self.max_seq_len_cached = None
+        else:
+            self.register_buffer('inv_freq', inv_freq)
+            self.max_seq_len_cached = None
+            self.cos_cached = None
+            self.sin_cached = None
+        self.precision = precision
+
+    def _load_from_state_dict(self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys,
+                              error_msgs):
+        pass
+
+    def forward(self, x, seq_dim=1, seq_len=None):
+        if seq_len is None:
+            seq_len = x.shape[seq_dim]
+        if self.max_seq_len_cached is None or (seq_len > self.max_seq_len_cached):
+            self.max_seq_len_cached = None if self.learnable else seq_len
+            t = torch.arange(seq_len, device=x.device, dtype=self.inv_freq.dtype)
+            freqs = torch.einsum('i,j->ij', t, self.inv_freq)
+            # Different from paper, but it uses a different permutation in order to obtain the same calculation
+            emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
+            if self.precision == torch.bfloat16:
+                emb = emb.float()
+
+            # [sx, 1 (b * np), hn]
+            cos_cached = emb.cos()[:, None, :]
+            sin_cached = emb.sin()[:, None, :]
+            if self.precision == torch.bfloat16:
+                cos_cached = cos_cached.bfloat16()
+                sin_cached = sin_cached.bfloat16()
+            if self.learnable:
+                return cos_cached, sin_cached
+            self.cos_cached, self.sin_cached = cos_cached, sin_cached
+        return self.cos_cached[:seq_len, ...], self.sin_cached[:seq_len, ...]
+
+
+def rotate_half(x):
+    x1, x2 = x[..., :x.shape[-1] // 2], x[..., x.shape[-1] // 2:]
+    return torch.cat((-x2, x1), dim=x1.ndim - 1)  # dim=-1 triggers a bug in earlier torch versions
+
+
+@torch.jit.script
+def apply_rotary_pos_emb_index(q, k, cos, sin, position_id):
+    # position_id: [sq, b], q, k: [sq, b, np, hn], cos: [sq, 1, hn] -> [sq, b, 1, hn]
+    cos, sin = F.embedding(position_id, cos.squeeze(1)).unsqueeze(2), \
+        F.embedding(position_id, sin.squeeze(1)).unsqueeze(2)
+    q, k = (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)
+    return q, k
+
+
+def attention_fn(
+        self,
+        query_layer,
+        key_layer,
+        value_layer,
+        attention_mask,
+        hidden_size_per_partition,
+        layer_id,
+        layer_past=None,
+        scaling_attention_score=True,
+        use_cache=False,
+):
+    if layer_past is not None:
+        past_key, past_value = layer_past
+        key_layer = torch.cat((past_key, key_layer), dim=0)
+        value_layer = torch.cat((past_value, value_layer), dim=0)
+
+    # seqlen, batch, num_attention_heads, hidden_size_per_attention_head
+    seq_len, b, nh, hidden_size = key_layer.shape
+
+    if use_cache:
+        present = (key_layer, value_layer)
+    else:
+        present = None
+
+    query_key_layer_scaling_coeff = float(layer_id + 1)
+    if scaling_attention_score:
+        query_layer = query_layer / (math.sqrt(hidden_size) * query_key_layer_scaling_coeff)
+
+    # ===================================
+    # Raw attention scores. [b, np, s, s]
+    # ===================================
+
+    # [b, np, sq, sk]
+    output_size = (query_layer.size(1), query_layer.size(2), query_layer.size(0), key_layer.size(0))
+
+    # [sq, b, np, hn] -> [sq, b * np, hn]
+    query_layer = query_layer.view(output_size[2], output_size[0] * output_size[1], -1)
+    # [sk, b, np, hn] -> [sk, b * np, hn]
+    key_layer = key_layer.view(output_size[3], output_size[0] * output_size[1], -1)
+
+    matmul_result = torch.empty(
+        output_size[0] * output_size[1],
+        output_size[2],
+        output_size[3],
+        dtype=query_layer.dtype,
+        device=query_layer.device,
+    )
+
+    matmul_result = torch.baddbmm(
+        matmul_result,
+        query_layer.transpose(0, 1),  # [b * np, sq, hn]
+        key_layer.transpose(0, 1).transpose(1, 2),  # [b * np, hn, sk]
+        beta=0.0,
+        alpha=1.0,
+    )
+
+    # change view to [b, np, sq, sk]
+    attention_scores = matmul_result.view(*output_size)
+
+    if self.scale_mask_softmax:
+        self.scale_mask_softmax.scale = query_key_layer_scaling_coeff
+        attention_probs = self.scale_mask_softmax(attention_scores, attention_mask.contiguous())
+    else:
+        if not (attention_mask == 0).all():
+            # if auto-regressive, skip
+            attention_scores.masked_fill_(attention_mask, -10000.0)
+        dtype = attention_scores.type()
+        attention_scores = attention_scores.float()
+        attention_scores = attention_scores * query_key_layer_scaling_coeff
+
+        attention_probs = F.softmax(attention_scores, dim=-1)
+
+        attention_probs = attention_probs.type(dtype)
+
+    # =========================
+    # Context layer. [sq, b, hp]
+    # =========================
+
+    # value_layer -> context layer.
+    # [sk, b, np, hn] --> [b, np, sq, hn]
+
+    # context layer shape: [b, np, sq, hn]
+    output_size = (value_layer.size(1), value_layer.size(2), query_layer.size(0), value_layer.size(3))
+
+    # change view [sk, b * np, hn]
+    value_layer = value_layer.view(value_layer.size(0), output_size[0] * output_size[1], -1)
+
+    # change view [b * np, sq, sk]
+    attention_probs = attention_probs.view(output_size[0] * output_size[1], output_size[2], -1)
+
+    # matmul: [b * np, sq, hn]
+    context_layer = torch.bmm(attention_probs, value_layer.transpose(0, 1))
+
+    # change view [b, np, sq, hn]
+    context_layer = context_layer.view(*output_size)
+
+    # [b, np, sq, hn] --> [sq, b, np, hn]
+    context_layer = context_layer.permute(2, 0, 1, 3).contiguous()
+
+    # [sq, b, np, hn] --> [sq, b, hp]
+    new_context_layer_shape = context_layer.size()[:-2] + (hidden_size_per_partition,)
+    context_layer = context_layer.view(*new_context_layer_shape)
+
+    outputs = (context_layer, present, attention_probs)
+
+    return outputs
+
+
+class SelfAttention(torch.nn.Module):
+    def __init__(self, hidden_size, num_attention_heads,
+                 layer_id, hidden_size_per_attention_head=None, bias=True,
+                 params_dtype=torch.float, position_encoding_2d=True):
+        super(SelfAttention, self).__init__()
+
+        self.layer_id = layer_id
+        self.hidden_size = hidden_size
+        self.hidden_size_per_partition = hidden_size
+        self.num_attention_heads = num_attention_heads
+        self.num_attention_heads_per_partition = num_attention_heads
+        self.position_encoding_2d = position_encoding_2d
+        self.rotary_emb = RotaryEmbedding(
+            self.hidden_size // (self.num_attention_heads * 2)
+            if position_encoding_2d
+            else self.hidden_size // self.num_attention_heads,
+            base=10000,
+            precision=torch.half,
+            learnable=False,
+        )
+
+        self.scale_mask_softmax = None
+
+        if hidden_size_per_attention_head is None:
+            self.hidden_size_per_attention_head = hidden_size // num_attention_heads
+        else:
+            self.hidden_size_per_attention_head = hidden_size_per_attention_head
+
+        self.inner_hidden_size = num_attention_heads * self.hidden_size_per_attention_head
+
+        # Strided linear layer.
+        self.query_key_value = skip_init(
+            torch.nn.Linear,
+            hidden_size,
+            3 * self.inner_hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+
+        self.dense = skip_init(
+            torch.nn.Linear,
+            self.inner_hidden_size,
+            hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+
+    @staticmethod
+    def attention_mask_func(attention_scores, attention_mask):
+        attention_scores.masked_fill_(attention_mask, -10000.0)
+        return attention_scores
+
+    def split_tensor_along_last_dim(self, tensor, num_partitions,
+                                    contiguous_split_chunks=False):
+        """Split a tensor along its last dimension.
+        Arguments:
+            tensor: input tensor.
+            num_partitions: number of partitions to split the tensor
+            contiguous_split_chunks: If True, make each chunk contiguous
+                                    in memory.
+        """
+        # Get the size and dimension.
+        last_dim = tensor.dim() - 1
+        last_dim_size = tensor.size()[last_dim] // num_partitions
+        # Split.
+        tensor_list = torch.split(tensor, last_dim_size, dim=last_dim)
+        # Note: torch.split does not create contiguous tensors by default.
+        if contiguous_split_chunks:
+            return tuple(chunk.contiguous() for chunk in tensor_list)
+
+        return tensor_list
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            position_ids,
+            attention_mask: torch.Tensor,
+            layer_id,
+            layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+            use_cache: bool = False,
+            output_attentions: bool = False,
+    ):
+        """
+        hidden_states: [seq_len, batch, hidden_size]
+        attention_mask: [(1, 1), seq_len, seq_len]
+        """
+
+        # [seq_len, batch, 3 * hidden_size]
+        mixed_raw_layer = self.query_key_value(hidden_states)
+
+        # [seq_len, batch, 3 * hidden_size] --> [seq_len, batch, num_attention_heads, 3 * hidden_size_per_attention_head]
+        new_tensor_shape = mixed_raw_layer.size()[:-1] + (
+            self.num_attention_heads_per_partition,
+            3 * self.hidden_size_per_attention_head,
+        )
+        mixed_raw_layer = mixed_raw_layer.view(*new_tensor_shape)
+
+        # [seq_len, batch, num_attention_heads, hidden_size_per_attention_head]
+        (query_layer, key_layer, value_layer) = self.split_tensor_along_last_dim(mixed_raw_layer, 3)
+
+        if self.position_encoding_2d:
+            q1, q2 = query_layer.chunk(2, dim=(query_layer.ndim - 1))
+            k1, k2 = key_layer.chunk(2, dim=(key_layer.ndim - 1))
+            cos, sin = self.rotary_emb(q1, seq_len=position_ids.max() + 1)
+            position_ids, block_position_ids = position_ids[:, 0, :].transpose(0, 1).contiguous(), \
+                position_ids[:, 1, :].transpose(0, 1).contiguous()
+            q1, k1 = apply_rotary_pos_emb_index(q1, k1, cos, sin, position_ids)
+            q2, k2 = apply_rotary_pos_emb_index(q2, k2, cos, sin, block_position_ids)
+            query_layer = torch.concat([q1, q2], dim=(q1.ndim - 1))
+            key_layer = torch.concat([k1, k2], dim=(k1.ndim - 1))
+        else:
+            position_ids = position_ids.transpose(0, 1)
+            cos, sin = self.rotary_emb(value_layer, seq_len=position_ids.max() + 1)
+            # [seq_len, batch, num_attention_heads, hidden_size_per_attention_head]
+            query_layer, key_layer = apply_rotary_pos_emb_index(query_layer, key_layer, cos, sin, position_ids)
+
+        # [seq_len, batch, hidden_size]
+        context_layer, present, attention_probs = attention_fn(
+            self=self,
+            query_layer=query_layer,
+            key_layer=key_layer,
+            value_layer=value_layer,
+            attention_mask=attention_mask,
+            hidden_size_per_partition=self.hidden_size_per_partition,
+            layer_id=layer_id,
+            layer_past=layer_past,
+            use_cache=use_cache
+        )
+
+        output = self.dense(context_layer)
+
+        outputs = (output, present)
+
+        if output_attentions:
+            outputs += (attention_probs,)
+
+        return outputs  # output, present, attention_probs
+
+
+class GEGLU(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.activation_fn = F.gelu
+
+    def forward(self, x):
+        # dim=-1 breaks in jit for pt<1.10
+        x1, x2 = x.chunk(2, dim=(x.ndim - 1))
+        return x1 * self.activation_fn(x2)
+
+
+class GLU(torch.nn.Module):
+    def __init__(self, hidden_size, inner_hidden_size=None,
+                 layer_id=None, bias=True, activation_func=gelu, params_dtype=torch.float):
+        super(GLU, self).__init__()
+        self.layer_id = layer_id
+        self.activation_func = activation_func
+
+        # Project to 4h.
+        self.hidden_size = hidden_size
+        if inner_hidden_size is None:
+            inner_hidden_size = 4 * hidden_size
+        self.inner_hidden_size = inner_hidden_size
+        self.dense_h_to_4h = skip_init(
+            torch.nn.Linear,
+            self.hidden_size,
+            self.inner_hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+        # Project back to h.
+        self.dense_4h_to_h = skip_init(
+            torch.nn.Linear,
+            self.inner_hidden_size,
+            self.hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+
+    def forward(self, hidden_states):
+        """
+        hidden_states: [seq_len, batch, hidden_size]
+        """
+
+        # [seq_len, batch, inner_hidden_size]
+        intermediate_parallel = self.dense_h_to_4h(hidden_states)
+
+        intermediate_parallel = self.activation_func(intermediate_parallel)
+
+        output = self.dense_4h_to_h(intermediate_parallel)
+
+        return output
+
+
+class GLMBlock(torch.nn.Module):
+    def __init__(
+            self,
+            hidden_size,
+            num_attention_heads,
+            layernorm_epsilon,
+            layer_id,
+            inner_hidden_size=None,
+            hidden_size_per_attention_head=None,
+            layernorm=LayerNorm,
+            use_bias=True,
+            params_dtype=torch.float,
+            num_layers=28,
+            position_encoding_2d=True
+    ):
+        super(GLMBlock, self).__init__()
+        # Set output layer initialization if not provided.
+
+        self.layer_id = layer_id
+
+        # Layernorm on the input data.
+        self.input_layernorm = layernorm(hidden_size, eps=layernorm_epsilon)
+
+        self.position_encoding_2d = position_encoding_2d
+
+        # Self attention.
+        self.attention = SelfAttention(
+            hidden_size,
+            num_attention_heads,
+            layer_id,
+            hidden_size_per_attention_head=hidden_size_per_attention_head,
+            bias=use_bias,
+            params_dtype=params_dtype,
+            position_encoding_2d=self.position_encoding_2d
+        )
+
+        # Layernorm on the input data.
+        self.post_attention_layernorm = layernorm(hidden_size, eps=layernorm_epsilon)
+
+        self.num_layers = num_layers
+
+        # GLU
+        self.mlp = GLU(
+            hidden_size,
+            inner_hidden_size=inner_hidden_size,
+            bias=use_bias,
+            layer_id=layer_id,
+            params_dtype=params_dtype,
+        )
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            position_ids,
+            attention_mask: torch.Tensor,
+            layer_id,
+            layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+            use_cache: bool = False,
+            output_attentions: bool = False,
+    ):
+        """
+        hidden_states: [seq_len, batch, hidden_size]
+        attention_mask: [(1, 1), seq_len, seq_len]
+        """
+
+        # Layer norm at the begining of the transformer layer.
+        # [seq_len, batch, hidden_size]
+        attention_input = self.input_layernorm(hidden_states)
+
+        # Self attention.
+        attention_outputs = self.attention(
+            attention_input,
+            position_ids,
+            attention_mask=attention_mask,
+            layer_id=layer_id,
+            layer_past=layer_past,
+            use_cache=use_cache,
+            output_attentions=output_attentions
+        )
+
+        attention_output = attention_outputs[0]
+
+        outputs = attention_outputs[1:]
+
+        # Residual connection.
+        alpha = (2 * self.num_layers) ** 0.5
+        hidden_states = attention_input * alpha + attention_output
+
+        mlp_input = self.post_attention_layernorm(hidden_states)
+
+        # MLP.
+        mlp_output = self.mlp(mlp_input)
+
+        # Second residual connection.
+        output = mlp_input * alpha + mlp_output
+
+        if use_cache:
+            outputs = (output,) + outputs
+        else:
+            outputs = (output,) + outputs[1:]
+
+        return outputs  # hidden_states, present, attentions
+
+
+class ChatGLMPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and
+    a simple interface for downloading and loading pretrained models.
+    """
+
+    is_parallelizable = True
+    supports_gradient_checkpointing = False
+    config_class = ChatGLMConfig
+    base_model_prefix = "transformer"
+    _no_split_modules = ["GLM6BBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module: nn.Module):
+        """Initialize the weights."""
+        return
+
+
+CHATGLM_6B_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general
+    usage and behavior.
+
+    Parameters:
+        config ([`~ChatGLM6BConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the configuration.
+            Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+CHATGLM_6B_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the vocabulary.
+
+            Indices can be obtained using [`ChatGLM6BTokenizer`].
+            See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+        token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`:
+
+            - 0 corresponds to a *sentence A* token,
+            - 1 corresponds to a *sentence B* token.
+
+            [What are token type IDs?](../glossary#token-type-ids)
+        position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings.
+            Selected in the range `[0, config.max_position_embeddings - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
+            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
+            This is useful if you want more control over how to convert *input_ids* indices into associated vectors
+            than the model's internal embedding lookup matrix.
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare ChatGLM-6B Model transformer outputting raw hidden-states without any specific head on top.",
+    CHATGLM_6B_START_DOCSTRING,
+)
+class ChatGLMModel(ChatGLMPreTrainedModel):
+    """
+
+    The model can behave as an encoder (with only self-attention) as well
+    as a decoder, in which case a layer of cross-attention is added between
+    the self-attention layers, following the architecture described in [Attention is
+    all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani,
+    Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+
+    To behave as an decoder the model needs to be initialized with the
+    `is_decoder` argument of the configuration set to `True`.
+    To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder`
+    argument and `add_cross_attention` set to `True`; an
+    `encoder_hidden_states` is then expected as an input to the forward pass.
+    """
+
+    def __init__(self, config: ChatGLMConfig):
+        super().__init__(config)
+
+        # recording parameters
+        self.max_sequence_length = config.max_sequence_length
+        self.hidden_size = config.hidden_size
+        self.params_dtype = torch.half
+        self.num_attention_heads = config.num_attention_heads
+        self.vocab_size = config.vocab_size
+        self.num_layers = config.num_layers
+        self.layernorm_epsilon = config.layernorm_epsilon
+        self.inner_hidden_size = config.inner_hidden_size
+        self.hidden_size_per_attention_head = self.hidden_size // self.num_attention_heads
+        self.position_encoding_2d = config.position_encoding_2d
+
+        self.word_embeddings = skip_init(
+            torch.nn.Embedding,
+            num_embeddings=self.vocab_size, embedding_dim=self.hidden_size,
+            dtype=self.params_dtype
+        )
+
+        def get_layer(layer_id):
+            return GLMBlock(
+                self.hidden_size,
+                self.num_attention_heads,
+                self.layernorm_epsilon,
+                layer_id,
+                inner_hidden_size=self.inner_hidden_size,
+                hidden_size_per_attention_head=self.hidden_size_per_attention_head,
+                layernorm=LayerNorm,
+                use_bias=True,
+                params_dtype=self.params_dtype,
+                position_encoding_2d=self.position_encoding_2d,
+            )
+
+        self.layers = torch.nn.ModuleList(
+            [get_layer(layer_id) for layer_id in range(self.num_layers)]
+        )
+
+        # Final layer norm before output.
+        self.final_layernorm = LayerNorm(self.hidden_size, eps=self.layernorm_epsilon)
+
+    def get_input_embeddings(self):
+        return self.word_embeddings
+
+    def set_input_embeddings(self, new_embeddings: torch.Tensor):
+        self.word_embeddings = new_embeddings
+
+    @staticmethod
+    def get_masks(seq, device):
+        context_length = seq.index(150004) + 1
+
+        attention_mask = torch.ones((1, len(seq), len(seq)), device=device)
+        attention_mask.tril_()
+        attention_mask[..., :context_length - 1] = 1
+        attention_mask.unsqueeze_(1)
+        attention_mask = (attention_mask < 0.5).bool()
+
+        return attention_mask
+
+    def get_position_ids(self, seq, mask_position, device, gmask=False):
+        context_length = seq.index(150004) + 1
+        if self.position_encoding_2d:
+            seq_length = seq.index(150004)
+            position_ids = torch.arange(context_length, dtype=torch.long, device=device)
+            if not gmask:
+                position_ids[seq_length:] = mask_position
+            block_position_ids = torch.cat((
+                torch.zeros(seq_length, dtype=torch.long, device=device),
+                torch.arange(context_length - seq_length, dtype=torch.long, device=device) + 1
+            ))
+            position_ids = torch.stack((position_ids, block_position_ids), dim=0)
+        else:
+            position_ids = torch.arange(context_length, dtype=torch.long, device=device)
+            if not gmask:
+                position_ids[context_length - 1:] = mask_position
+
+        position_ids = position_ids.unsqueeze(0)
+
+        return position_ids
+
+    @add_start_docstrings_to_model_forward(CHATGLM_6B_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPastAndCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
+            inputs_embeds: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor, ...], BaseModelOutputWithPast]:
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape[:2]
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape[:2]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        if past_key_values is None:
+            past_key_values = tuple([None] * len(self.layers))
+
+            MASK, gMASK = 150000, 150001
+            mask_token = MASK if MASK in input_ids else gMASK
+            use_gmask = False if MASK in input_ids else gMASK
+            seq = input_ids[0].tolist()
+
+            mask_position = seq.index(mask_token)
+
+            if attention_mask is None:
+                attention_mask = self.get_masks(
+                    seq=seq,
+                    device=input_ids.device
+                )
+
+            if position_ids is None:
+                position_ids = self.get_position_ids(
+                    seq=seq,
+                    mask_position=mask_position,
+                    device=input_ids.device,
+                    gmask=use_gmask
+                )
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+
+        # [seq_len, batch, hidden_size]
+        hidden_states = inputs_embeds.transpose(0, 1)
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_hidden_states = () if output_hidden_states else None
+
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+        if past_key_values[0] is not None:
+            past_key_values_length = past_key_values[0][0].shape[0]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+        if attention_mask is None:
+            attention_mask = torch.zeros(1, 1, device=input_ids.device).bool()
+
+        else:
+            attention_mask = attention_mask.to(input_ids.device)
+
+        for i, layer in enumerate(self.layers):
+
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_ret = layer(
+                hidden_states,
+                position_ids=position_ids,
+                attention_mask=attention_mask,
+                layer_id=torch.tensor(i),
+                layer_past=past_key_values[i],
+                use_cache=use_cache,
+                output_attentions=output_attentions
+            )
+
+            hidden_states = layer_ret[0]
+
+            if use_cache:
+                presents = presents + (layer_ret[1],)
+
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_ret[2 if use_cache else 1],)
+
+        # Final layer norm.
+        hidden_states = self.final_layernorm(hidden_states)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+class ChatGLMForConditionalGeneration(ChatGLMPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        # self.hidden_size = config.hidden_size
+        # self.params_dtype = torch.half
+        # self.vocab_size = config.vocab_size
+        self.max_sequence_length = config.max_sequence_length
+
+        self.position_encoding_2d = config.position_encoding_2d
+
+        self.transformer = ChatGLMModel(config)
+
+        self.lm_head = skip_init(
+            nn.Linear,
+            config.hidden_size,
+            config.vocab_size,
+            bias=False,
+            dtype=torch.half
+        )
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def get_masks_and_position_ids(self, seq, mask_position, context_length, device, gmask=False):
+        attention_mask = torch.ones((1, context_length, context_length), device=device)
+        attention_mask.tril_()
+        attention_mask[..., :context_length - 1] = 1
+        attention_mask.unsqueeze_(1)
+        attention_mask = (attention_mask < 0.5).bool()
+
+        if self.position_encoding_2d:
+            seq_length = seq.index(150004)
+            position_ids = torch.arange(context_length, dtype=torch.long, device=device)
+            if not gmask:
+                position_ids[seq_length:] = mask_position
+            block_position_ids = torch.cat((
+                torch.zeros(seq_length, dtype=torch.long, device=device),
+                torch.arange(context_length - seq_length, dtype=torch.long, device=device) + 1
+            ))
+            position_ids = torch.stack((position_ids, block_position_ids), dim=0)
+        else:
+            position_ids = torch.arange(context_length, dtype=torch.long, device=device)
+            if not gmask:
+                position_ids[context_length - 1:] = mask_position
+
+        position_ids = position_ids.unsqueeze(0)
+
+        return attention_mask, position_ids
+
+    def prepare_inputs_for_generation(
+            self,
+            input_ids: torch.LongTensor,
+            past: Optional[torch.Tensor] = None,
+            past_key_values: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            **kwargs
+    ) -> dict:
+
+        MASK, gMASK = 150000, 150001
+        mask_token = MASK if MASK in input_ids else gMASK
+        use_gmask = False if MASK in input_ids else gMASK
+        seq = input_ids[0].tolist()
+        mask_position = seq.index(mask_token)
+
+        if mask_token not in seq:
+            raise ValueError("You have to add either [MASK] or [gMASK] in your input")
+
+        # only last token for input_ids if past is not None
+        if past is not None or past_key_values is not None:
+            context_length = seq.index(150004)
+            last_token = input_ids[:, -1].unsqueeze(-1)
+            if self.position_encoding_2d:
+                position_ids = torch.tensor([[[mask_position], [len(seq) - context_length]]], dtype=torch.long,
+                                            device=input_ids.device)
+            else:
+                position_ids = torch.tensor([[mask_position]], dtype=torch.long, device=input_ids.device)
+
+            if past is None:
+                past = past_key_values
+            return {
+                "input_ids": last_token,
+                "past_key_values": past,
+                "position_ids": position_ids,
+            }
+        else:
+            attention_mask, position_ids = self.get_masks_and_position_ids(
+                seq=seq,
+                mask_position=mask_position,
+                context_length=len(seq),
+                device=input_ids.device,
+                gmask=use_gmask
+            )
+
+            return {
+                "input_ids": input_ids,
+                "past_key_values": past,
+                "position_ids": position_ids,
+                "attention_mask": attention_mask
+            }
+
+    def forward(
+            self,
+            input_ids: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            past_key_values: Optional[Tuple[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.Tensor] = None,
+            labels: Optional[torch.Tensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ):
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.transformer(
+            input_ids=input_ids,
+            position_ids=position_ids,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = transformer_outputs[0]
+
+        lm_logits = self.lm_head(hidden_states).permute(1, 0, 2).contiguous()
+
+        loss = None
+        if labels is not None:
+            lm_logits = lm_logits.to(torch.float32)
+
+            # Shift so that tokens < n predict n
+            shift_logits = lm_logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+            lm_logits = lm_logits.to(hidden_states.dtype)
+            loss = loss.to(hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+            past: Tuple[Tuple[torch.Tensor, torch.Tensor], ...], beam_idx: torch.LongTensor
+    ) -> Tuple[Tuple[torch.Tensor, torch.Tensor], ...]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
+        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+
+        Output shares the same memory storage as `past`.
+        """
+        return tuple(
+            (
+                layer_past[0].index_select(1, beam_idx.to(layer_past[0].device)),
+                layer_past[1].index_select(1, beam_idx.to(layer_past[1].device)),
+            )
+            for layer_past in past
+        )
+
+    @torch.no_grad()
+    def chat(self, tokenizer, query: str, history: List[Tuple[str, str]] = None, max_length: int = 2048, num_beams=1,
+             do_sample=True, top_p=0.7, temperature=0.95, **kwargs):
+        if history is None:
+            history = []
+        gen_kwargs = {"max_length": max_length, "num_beams": num_beams, "do_sample": do_sample, "top_p": top_p,
+                      "temperature": temperature, **kwargs}
+        if not history:
+            prompt = query
+        else:
+            prompt = ""
+            for i, (old_query, response) in enumerate(history):
+                prompt += "[Round {}]\n问：{}\n答：{}\n".format(i, old_query, response)
+            prompt += "[Round {}]\n问：{}\n答：".format(len(history), query)
+        input_ids = tokenizer([prompt], return_tensors="pt", padding=True)
+        input_ids = input_ids.to(self.device)
+        outputs = self.generate(**input_ids, **gen_kwargs)
+        outputs = outputs.tolist()[0][len(input_ids["input_ids"][0]) - 2:]
+        response = tokenizer.decode(outputs)
+        response = response.strip()
+        response = response.replace("[[训练时间]]", "2023年")
+        history = history + [(query, response)]
+        return response, history
+
+    @torch.no_grad()
+    def generate(
+            self,
+            **kwargs,
+    ):
+        MASK, gMASK = 150000, 150001
+        bos, eos = 150004, 150005
+
+        if "eos_token_id" not in kwargs:
+            kwargs["eos_token_id"] = eos
+
+        stop = False
+
+        return_seqs = []
+
+        while True:
+            output_ids = super().generate(**kwargs)
+
+            return_seqs = []
+            max_length = 0
+
+            for i in range(output_ids.shape[0]):
+                output_seq = output_ids[i].tolist()
+                mask_token = MASK if MASK in output_seq else gMASK
+                mask_position = output_seq.index(mask_token)
+                bos_position = output_seq.index(bos)
+                if eos in output_seq:
+                    eos_position = output_seq.index(eos)
+                else:
+                    eos_position = len(output_seq)
+
+                return_seq = output_seq[:mask_position] + output_seq[bos_position + 1:eos_position] + output_seq[
+                                                                                                      mask_position + 1:bos_position]
+                max_length = max(max_length, len(return_seq))
+                return_seqs.append(return_seq)
+
+            for i in range(output_ids.shape[0]):
+                return_seqs[i] = [0] * (max_length - len(return_seqs[i])) + return_seqs[i]  # padding
+                if mask_token not in return_seqs[i]:
+                    stop = True
+
+            if stop:
+                break
+
+            for return_seq in return_seqs:
+                return_seq += [bos]
+
+            kwargs['input_ids'] = torch.tensor(return_seqs, dtype=torch.long, device=kwargs['input_ids'].device)
+
+        return torch.tensor(return_seqs, dtype=torch.long, device=kwargs['input_ids'].device)
+
+    def quantize(self, bits: int):
+        from .quantization import quantize
+        self.transformer = quantize(self.transformer, bits)
+        return self
diff --git a/moe/pytorch_model.bin.index.json b/moe/pytorch_model.bin.index.json
new file mode 100644
index 0000000000000000000000000000000000000000..b8ada2bdf39c8297dc2b3159270227c587bd13e9
--- /dev/null
+++ b/moe/pytorch_model.bin.index.json
@@ -0,0 +1,375 @@
+{
+  "metadata": {
+    "total_size": 13744473856
+  },
+  "weight_map": {
+    "lm_head.weight": "pytorch_model-00008-of-00008.bin",
+    "transformer.final_layernorm.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.final_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.0.attention.dense.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.attention.dense.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.attention.query_key_value.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.attention.query_key_value.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.attention.rotary_emb.inv_freq": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.input_layernorm.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.mlp.dense_4h_to_h.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.mlp.dense_4h_to_h.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.mlp.dense_h_to_4h.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.mlp.dense_h_to_4h.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.post_attention_layernorm.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.attention.dense.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.attention.dense.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.attention.query_key_value.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.attention.query_key_value.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.attention.rotary_emb.inv_freq": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.input_layernorm.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.1.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.1.mlp.dense_h_to_4h.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.mlp.dense_h_to_4h.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.post_attention_layernorm.bias": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+    "transformer.layers.10.attention.dense.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.attention.dense.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.post_attention_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.10.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.11.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.11.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.11.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.11.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.11.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.11.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.11.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.11.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.11.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.11.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.11.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.11.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.11.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.attention.query_key_value.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.attention.query_key_value.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.12.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.attention.query_key_value.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.attention.query_key_value.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.13.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.attention.query_key_value.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.attention.query_key_value.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.14.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.attention.query_key_value.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.attention.query_key_value.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.15.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.16.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.16.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.16.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "transformer.layers.16.mlp.dense_4h_to_h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.mlp.dense_4h_to_h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.16.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.attention.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.input_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.mlp.dense_4h_to_h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.mlp.dense_4h_to_h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.17.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.attention.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.input_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.mlp.dense_4h_to_h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.mlp.dense_4h_to_h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.18.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.attention.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.input_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.mlp.dense_4h_to_h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.mlp.dense_4h_to_h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.19.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.2.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.mlp.dense_h_to_4h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.mlp.dense_h_to_4h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.2.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.20.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.attention.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.input_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.20.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.20.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.20.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "transformer.layers.21.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.mlp.dense_h_to_4h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.mlp.dense_h_to_4h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.21.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.mlp.dense_h_to_4h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.mlp.dense_h_to_4h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.22.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.mlp.dense_h_to_4h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.mlp.dense_h_to_4h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.23.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.mlp.dense_h_to_4h.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.mlp.dense_h_to_4h.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.24.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.mlp.dense_4h_to_h.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.25.mlp.dense_4h_to_h.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.25.mlp.dense_h_to_4h.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.25.mlp.dense_h_to_4h.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.25.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.25.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "transformer.layers.26.attention.dense.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.attention.dense.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.attention.query_key_value.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.attention.query_key_value.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.attention.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.input_layernorm.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.mlp.dense_4h_to_h.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.mlp.dense_4h_to_h.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.mlp.dense_h_to_4h.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.mlp.dense_h_to_4h.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.post_attention_layernorm.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.26.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.attention.dense.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.attention.dense.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.attention.query_key_value.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.attention.query_key_value.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.attention.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.input_layernorm.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.mlp.dense_4h_to_h.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.mlp.dense_4h_to_h.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.mlp.dense_h_to_4h.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.mlp.dense_h_to_4h.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.post_attention_layernorm.bias": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.27.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "transformer.layers.3.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.mlp.dense_h_to_4h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.mlp.dense_h_to_4h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.3.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.mlp.dense_h_to_4h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.mlp.dense_h_to_4h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.4.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.mlp.dense_h_to_4h.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.mlp.dense_h_to_4h.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.5.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.6.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.6.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.6.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.6.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.6.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "transformer.layers.7.attention.dense.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.attention.dense.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.post_attention_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.7.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.attention.dense.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.attention.dense.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.post_attention_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.8.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.attention.dense.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.attention.dense.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.post_attention_layernorm.bias": "pytorch_model-00003-of-00008.bin",
+    "transformer.layers.9.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "transformer.word_embeddings.weight": "pytorch_model-00001-of-00008.bin"
+  }
+}
diff --git a/moe/quantization.py b/moe/quantization.py
new file mode 100644
index 0000000000000000000000000000000000000000..bf30790cd8cd4adbead4f8cb64afd6db7ff29cf2
--- /dev/null
+++ b/moe/quantization.py
@@ -0,0 +1,187 @@
+from torch.nn import Linear
+from torch.nn.parameter import Parameter
+
+import bz2
+import torch
+import base64
+import ctypes
+
+from typing import List
+from cpm_kernels.kernels.base import LazyKernelCModule, KernelFunction, round_up
+
+
+class W8A16Linear(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, inp: torch.Tensor, quant_w: torch.Tensor, scale_w: torch.Tensor, weight_bit_width):
+        ctx.inp_shape = inp.size()
+        ctx.weight_shape = quant_w.size()
+        ctx.weight_bit_width = weight_bit_width
+        out_features = quant_w.size(0)
+        inp = inp.contiguous().view(-1, inp.size(-1))
+        weight = extract_weight_to_half(quant_w, scale_w, weight_bit_width)
+        output = inp.mm(weight.t())
+        ctx.save_for_backward(inp, quant_w, scale_w)
+        return output.view(*(ctx.inp_shape[:-1] + (out_features,)))
+
+    @staticmethod
+    def backward(ctx, grad_output: torch.Tensor):
+        inp, quant_w, scale_w = ctx.saved_tensors
+        weight = extract_weight_to_half(quant_w, scale_w, ctx.weight_bit_width)
+        grad_output = grad_output.contiguous().view(-1, weight.size(0))
+        grad_input = grad_output.mm(weight)
+        grad_weight = grad_output.t().mm(inp)
+        return grad_input.view(ctx.inp_shape), grad_weight.view(ctx.weight_shape), None
+
+
+class Kernel:
+    def __init__(self, code: bytes, function_names: List[str]):
+        self.code = code
+        self._function_names = function_names
+        self._cmodule = LazyKernelCModule(self.code)
+
+        for name in self._function_names:
+            setattr(self, name, KernelFunction(self._cmodule, name))
+
+
+quantization_code = "$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"
+
+kernels = Kernel(
+    bz2.decompress(base64.b64decode(quantization_code)),
+    [
+        "int4WeightCompression",
+        "int4WeightExtractionFloat",
+        "int4WeightExtractionHalf",
+        "int8WeightExtractionFloat",
+        "int8WeightExtractionHalf",
+    ],
+)
+
+
+def compress_int4_weight(weight: torch.Tensor):  # (n, m)
+    with torch.cuda.device(weight.device):
+        n, m = weight.size(0), weight.size(1)
+        assert m % 2 == 0
+        m = m // 2
+        out = torch.empty(n, m, dtype=torch.int8, device="cuda")
+        stream = torch.cuda.current_stream()
+
+        gridDim = (n, 1, 1)
+        blockDim = (min(round_up(m, 32), 1024), 1, 1)
+
+        kernels.int4WeightCompression(
+            gridDim,
+            blockDim,
+            0,
+            stream,
+            [ctypes.c_void_p(weight.data_ptr()), ctypes.c_void_p(out.data_ptr()), ctypes.c_int32(n), ctypes.c_int32(m)],
+        )
+        return out
+
+
+def extract_weight_to_half(weight: torch.Tensor, scale_list: torch.Tensor, source_bit_width: int):
+    if source_bit_width == 8:
+        func = kernels.int8WeightExtractionHalf
+    elif source_bit_width == 4:
+        func = kernels.int4WeightExtractionHalf
+    else:
+        assert False, "Unsupported bit-width"
+
+    with torch.cuda.device(weight.device):
+        n, m = weight.size(0), weight.size(1)
+        out = torch.empty(n, m * (8 // source_bit_width), dtype=torch.half, device="cuda")
+        stream = torch.cuda.current_stream()
+
+        gridDim = (n, 1, 1)
+        blockDim = (min(round_up(m, 32), 1024), 1, 1)
+
+        func(
+            gridDim,
+            blockDim,
+            0,
+            stream,
+            [
+                ctypes.c_void_p(weight.data_ptr()),
+                ctypes.c_void_p(scale_list.data_ptr()),
+                ctypes.c_void_p(out.data_ptr()),
+                ctypes.c_int32(n),
+                ctypes.c_int32(m),
+            ],
+        )
+        return out
+
+
+class QuantizedLinear(Linear):
+    def __init__(self, weight_bit_width: int, weight_tensor=None, bias_tensor=None, *args, **kwargs):
+        super(QuantizedLinear, self).__init__(*args, **kwargs)
+        self.weight_bit_width = weight_bit_width
+
+        shape = self.weight.shape
+        del self.weight
+
+        if weight_tensor is None:
+            self.weight = torch.empty(
+                shape[0], shape[1] * weight_bit_width // 8, dtype=torch.int8, device=kwargs["device"]
+            )
+            self.weight_scale = torch.empty(shape[0], dtype=kwargs["params_dtype"], device=kwargs["device"])
+        else:
+            self.weight_scale = (weight_tensor.abs().max(dim=-1).values / ((2 ** (weight_bit_width - 1)) - 1)).half()
+            self.weight = torch.round(weight_tensor / self.weight_scale[:, None]).to(torch.int8)
+            if weight_bit_width == 4:
+                self.weight = compress_int4_weight(self.weight)
+
+        self.weight = Parameter(self.weight.to(kwargs["device"]), requires_grad=False)
+        self.weight_scale = Parameter(self.weight_scale.to(kwargs["device"]), requires_grad=False)
+        self.bias = Parameter(bias_tensor.to(kwargs["device"]), requires_grad=False)
+
+    def forward(self, input):
+        output = W8A16Linear.apply(input, self.weight, self.weight_scale, self.weight_bit_width)
+        if self.bias is not None:
+            output = output + self.bias
+        return output
+
+
+def quantize(model, weight_bit_width):
+    """Replace fp16 linear with quantized linear"""
+
+    for layer in model.layers:
+        layer.attention.query_key_value = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight_tensor=layer.attention.query_key_value.weight.to(torch.cuda.current_device()),
+            bias_tensor=layer.attention.query_key_value.bias,
+            in_features=layer.attention.query_key_value.in_features,
+            out_features=layer.attention.query_key_value.out_features,
+            bias=True,
+            dtype=torch.half,
+            device=layer.attention.query_key_value.weight.device,
+        )
+        layer.attention.dense = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight_tensor=layer.attention.dense.weight.to(torch.cuda.current_device()),
+            bias_tensor=layer.attention.dense.bias,
+            in_features=layer.attention.dense.in_features,
+            out_features=layer.attention.dense.out_features,
+            bias=True,
+            dtype=torch.half,
+            device=layer.attention.dense.weight.device,
+        )
+        layer.mlp.dense_h_to_4h = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight_tensor=layer.mlp.dense_h_to_4h.weight.to(torch.cuda.current_device()),
+            bias_tensor=layer.mlp.dense_h_to_4h.bias,
+            in_features=layer.mlp.dense_h_to_4h.in_features,
+            out_features=layer.mlp.dense_h_to_4h.out_features,
+            bias=True,
+            dtype=torch.half,
+            device=layer.mlp.dense_h_to_4h.weight.device,
+        )
+        layer.mlp.dense_4h_to_h = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight_tensor=layer.mlp.dense_4h_to_h.weight.to(torch.cuda.current_device()),
+            bias_tensor=layer.mlp.dense_4h_to_h.bias,
+            in_features=layer.mlp.dense_4h_to_h.in_features,
+            out_features=layer.mlp.dense_4h_to_h.out_features,
+            bias=True,
+            dtype=torch.half,
+            device=layer.mlp.dense_4h_to_h.weight.device,
+        )
+    return model
diff --git a/moe/temp1.wav b/moe/temp1.wav
new file mode 100644
index 0000000000000000000000000000000000000000..c8e988bfe54824c966d2238f847cbfc4a651c393
Binary files /dev/null and b/moe/temp1.wav differ
diff --git a/moe/temp2.wav b/moe/temp2.wav
new file mode 100644
index 0000000000000000000000000000000000000000..5f179652846abbaf12112c754fa42e2783ad8c0b
Binary files /dev/null and b/moe/temp2.wav differ
diff --git a/moe/tokenization_chatglm.py b/moe/tokenization_chatglm.py
new file mode 100644
index 0000000000000000000000000000000000000000..619a8c7a50fcd7369db05d33e0e4661e10c5ebe1
--- /dev/null
+++ b/moe/tokenization_chatglm.py
@@ -0,0 +1,345 @@
+"""Tokenization classes for ChatGLM."""
+import sys
+import unicodedata
+from typing import List, Optional, Union
+from functools import lru_cache
+import os
+import collections
+import re
+
+from transformers.tokenization_utils import PreTrainedTokenizer
+from icetk.text_tokenizer import TextTokenizer
+from icetk.utils import auto_create
+import icetk.sentencepiece_model_pb2 as sp_model
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "THUDM/chatglm-6b": 2048,
+}
+
+
+class SPTokenizer:
+    def __init__(
+        self,
+        vocab_file,
+        max_blank_length=80,
+        byte_fallback=True,
+    ):
+        assert vocab_file is not None
+        self.vocab_file = vocab_file
+        self.special_tokens = ["[MASK]", "[gMASK]", "[sMASK]", "<unused_0>", "<sop>", "<eop>", "<ENC>", "<dBLOCK>"]
+        self.max_blank_length = max_blank_length
+        self.byte_fallback = byte_fallback
+        self.text_tokenizer = self._build_text_tokenizer(encode_special_tokens=False)
+        self.special_text_tokenizer = self._build_text_tokenizer(encode_special_tokens=True)
+
+    @staticmethod
+    def _configure_tokenizer(
+        text_tokenizer: TextTokenizer,
+        special_tokens: List[str],
+        max_blank_length: int,
+        byte_fallback: bool,
+        encode_special_tokens=False,
+    ):
+        # special token
+        special_token_type = 4 if encode_special_tokens else 3  # 3 - CONTROL, 4 - USER_DEFINE
+        for token in special_tokens:
+            text_tokenizer.proto.pieces.append(
+                sp_model.ModelProto.SentencePiece(piece=token, score=0.0, type=special_token_type)
+            )
+        # whitespaces
+        for token in [SPTokenizer.get_tab_token()] + [
+            SPTokenizer.get_blank_token(i) for i in range(2, max_blank_length + 1)
+        ]:
+            text_tokenizer.proto.pieces.append(sp_model.ModelProto.SentencePiece(piece=token, score=0.0, type=4))
+        # byte fallback
+        if byte_fallback:
+            text_tokenizer.proto.trainer_spec.byte_fallback = True
+            for i in range(256):
+                text_tokenizer.proto.pieces.append(
+                    sp_model.ModelProto.SentencePiece(piece="<0x{:02X}>".format(i), score=0.0, type=6)
+                )
+        text_tokenizer.refresh()
+
+    def _build_text_tokenizer(self, encode_special_tokens=False):
+        tokenizer = TextTokenizer(self.vocab_file)
+        self._configure_tokenizer(
+            tokenizer, self.special_tokens, self.max_blank_length, self.byte_fallback, encode_special_tokens
+        )
+        return tokenizer
+
+    def _get_text_tokenizer(self, encode_special_tokens=False):
+        if encode_special_tokens:
+            return self.special_text_tokenizer
+        else:
+            return self.text_tokenizer
+
+    @staticmethod
+    def get_blank_token(length: int):
+        assert length >= 2
+        return f"<|blank_{length}|>"
+
+    @staticmethod
+    def get_tab_token():
+        return f"<|tab|>"
+
+    @property
+    def num_image_tokens(self):
+        return 20000
+
+    @property
+    def num_text_tokens(self):
+        return self.text_tokenizer.num_tokens
+
+    @property
+    def num_tokens(self):
+        return self.num_image_tokens + self.num_text_tokens
+
+    @staticmethod
+    def _encode_whitespaces(text: str, max_len: int = 80):
+        text = text.replace("\t", SPTokenizer.get_tab_token())
+        for i in range(max_len, 1, -1):
+            text = text.replace(" " * i, SPTokenizer.get_blank_token(i))
+        return text
+
+    def _preprocess(self, text: str, linebreak=True, whitespaces=True):
+        if linebreak:
+            text = text.replace("\n", "<n>")
+        if whitespaces:
+            text = self._encode_whitespaces(text, max_len=self.max_blank_length)
+        return text
+
+    def encode(
+        self, text: str, linebreak=True, whitespaces=True, special_tokens=False, add_dummy_prefix=True
+    ) -> List[int]:
+        """
+        @param text: Text to encode.
+        @param linebreak: Whether to encode newline (\n) in text.
+        @param whitespaces: Whether to encode multiple whitespaces or tab in text, useful for source code encoding.
+        @param special_tokens: Whether to encode special token ([MASK], [gMASK], etc.) in text.
+        @param add_dummy_prefix: Whether to add dummy blank space in the beginning.
+        """
+        text = self._preprocess(text, linebreak, whitespaces)
+        if not add_dummy_prefix:
+            text = "<n>" + text
+        tmp = self._get_text_tokenizer(encode_special_tokens=special_tokens).encode(text)
+        tokens = [x + self.num_image_tokens for x in tmp]
+        return tokens if add_dummy_prefix else tokens[2:]
+
+    def decode(self, text_ids: List[int], special_tokens=False) -> str:
+        ids = [int(_id) - self.num_image_tokens for _id in text_ids]
+        text = self._get_text_tokenizer(encode_special_tokens=special_tokens).decode(ids)
+        text = text.replace("<n>", "\n")
+        text = text.replace(SPTokenizer.get_tab_token(), "\t")
+        for i in range(2, self.max_blank_length + 1):
+            text = text.replace(self.get_blank_token(i), " " * i)
+        return text
+
+    def tokenize(
+        self, text: str, linebreak=True, whitespaces=True, special_tokens=False, add_dummy_prefix=True
+    ) -> List[str]:
+        """
+        @param text: Text to encode.
+        @param linebreak: Whether to encode newline (\n) in text.
+        @param whitespaces: Whether to encode multiple whitespaces or tab in text, useful for source code encoding.
+        @param special_tokens: Whether to encode special token ([MASK], [gMASK], etc.) in text.
+        @param add_dummy_prefix: Whether to add dummy blank space in the beginning.
+        """
+        text = self._preprocess(text, linebreak, whitespaces)
+        if not add_dummy_prefix:
+            text = "<n>" + text
+        tokens = self._get_text_tokenizer(encode_special_tokens=special_tokens).tokenize(text)
+        return tokens if add_dummy_prefix else tokens[2:]
+
+    def __getitem__(self, x: Union[int, str]):
+        if isinstance(x, int):
+            if x < self.num_image_tokens:
+                return "<image_{}>".format(x)
+            else:
+                return self.text_tokenizer.convert_id_to_token(x - self.num_image_tokens)
+        elif isinstance(x, str):
+            if x.startswith("<image_") and x.endswith(">") and x[7:-1].isdigit():
+                return int(x[7:-1])
+            else:
+                return self.text_tokenizer.convert_token_to_id(x) + self.num_image_tokens
+        else:
+            raise ValueError("The key should be str or int.")
+
+
+class ChatGLMTokenizer(PreTrainedTokenizer):
+    """
+    Construct a ChatGLM tokenizer. Based on byte-level Byte-Pair-Encoding.
+
+    Args:
+        vocab_file (`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = {"vocab_file": "ice_text.model"}
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["input_ids"]
+
+    def __init__(
+            self,
+            vocab_file,
+            do_lower_case=False,
+            remove_space=False,
+            bos_token='sop',
+            eos_token='eos',
+            eop_token='eop',
+            mask_token='[MASK]',
+            gmask_token='[gMASK]',
+            padding_side="left",
+            **kwargs
+    ) -> None:
+        super().__init__(
+            do_lower_case=do_lower_case,
+            remove_space=remove_space,
+            padding_side=padding_side,
+            **kwargs
+        )
+
+        self.do_lower_case = do_lower_case
+        self.remove_space = remove_space
+        self.vocab_file = vocab_file
+
+        self.bos_token = bos_token
+        self.eos_token = eos_token
+        self.eop_token = eop_token
+        self.mask_token = mask_token
+        self.gMASK_token = gmask_token
+
+        self.sp_tokenizer = SPTokenizer(vocab_file)
+
+        """ Initialisation """
+
+    @property
+    def eop_token_id(self) -> Optional[int]:
+        """
+        `Optional[int]`: Id of the end of sentence token in the vocabulary. Returns `None` if the token has not been
+        set.
+        """
+        if self.eop_token is None:
+            return None
+        return self.convert_tokens_to_ids(self.eop_token)
+
+    @property
+    def vocab_size(self):
+        """ Returns vocab size """
+        return self.sp_tokenizer.num_tokens
+
+    def get_vocab(self):
+        """ Returns vocab as a dict """
+        vocab = {self._convert_id_to_token(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def preprocess_text(self, inputs):
+        if self.remove_space:
+            outputs = " ".join(inputs.strip().split())
+        else:
+            outputs = inputs
+
+        if self.do_lower_case:
+            outputs = outputs.lower()
+
+        return outputs
+
+    def _tokenize(self, text, **kwargs):
+        """ Returns a tokenized string. """
+        text = self.preprocess_text(text)
+
+        seq = self.sp_tokenizer.tokenize(text)
+
+        return seq
+
+    def decode(
+            self,
+            token_ids: Union[List[int], List[List[int]]],
+            skip_special_tokens: bool = False,
+            clean_up_tokenization_spaces: bool = True,
+            spaces_between_special_tokens: bool = True,
+            **kwargs
+    ) -> str:
+        if isinstance(token_ids[0], list):
+            tokens = []
+            for single_token_ids in token_ids:
+                if self.pad_token_id in single_token_ids:  # remove pad
+                    single_token_ids = list(filter((self.pad_token_id).__ne__, single_token_ids))
+                tokens.append(self.sp_tokenizer.decode(single_token_ids))
+            return (tokens)
+        else:
+            if self.pad_token_id in token_ids:  # remove pad
+                token_ids = list(filter((self.pad_token_id).__ne__, token_ids))
+            return self.sp_tokenizer.decode(token_ids)
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str) in an id using the vocab. """
+        return self.sp_tokenizer[token]
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        return self.sp_tokenizer[index]
+
+    def save_vocabulary(self, save_directory, filename_prefix=None):
+        """
+        Save the vocabulary and special tokens file to a directory.
+
+        Args:
+            save_directory (`str`):
+                The directory in which to save the vocabulary.
+            filename_prefix (`str`, *optional*):
+                An optional prefix to add to the named of the saved files.
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        if os.path.isdir(save_directory):
+            vocab_file = os.path.join(
+                save_directory, VOCAB_FILES_NAMES["vocab_file"]
+            )
+        else:
+            vocab_file = save_directory
+
+        with open(self.vocab_file, 'rb') as fin:
+            proto_str = fin.read()
+
+        with open(vocab_file, "wb") as writer:
+            writer.write(proto_str)
+
+        return (vocab_file,)
+
+    def build_inputs_with_special_tokens(
+            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
+        adding special tokens. A BERT sequence has the following format:
+
+        - single sequence: `[CLS] X [SEP]`
+        - pair of sequences: `[CLS] A [SEP] B [SEP]`
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs to which the special tokens will be added.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
+        """
+        if token_ids_1 is not None:
+            token_ids_0 += token_ids_1
+        mask_ids = self.sp_tokenizer[self.mask_token]
+        gmask_ids = self.sp_tokenizer[self.gMASK_token]
+        if mask_ids not in token_ids_0 and gmask_ids not in token_ids_0:
+            token_ids_0 += [gmask_ids]
+
+        if token_ids_0[-1] != mask_ids and token_ids_0[-1] != gmask_ids:
+            token_ids_0 += [self.sp_tokenizer[self.eos_token]]
+
+        token_ids_0 += [self.sp_tokenizer[self.bos_token]]
+
+        return token_ids_0
diff --git a/moe/tokenizer_config.json b/moe/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..21c1748d14ccbb115d065fcf87007c0c0e6c098e
--- /dev/null
+++ b/moe/tokenizer_config.json
@@ -0,0 +1,19 @@
+{
+  "name_or_path": "THUDM/chatglm-6b",
+  "bos_token": "<sop>",
+  "eop_token": "<eop>",
+  "eos_token": "</s>",
+  "gmask_token": "[gMASK]",
+  "mask_token": "[MASK]",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>",
+  "remove_space": false,
+  "do_lower_case": false,
+  "tokenizer_class": "ChatGLMTokenizer",
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_chatglm.ChatGLMTokenizer",
+      null
+      ]
+  }
+}
diff --git a/output.wav b/output.wav
new file mode 100644
index 0000000000000000000000000000000000000000..5f179652846abbaf12112c754fa42e2783ad8c0b
Binary files /dev/null and b/output.wav differ
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2c2f4901eef9b73f7eebbcb737f98aa851f35e1d
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,24 @@
+Flask
+Cython==0.29.21
+librosa==0.8.0
+matplotlib==3.3.1
+numpy==1.21.6
+phonemizer==2.2.1
+scipy==1.5.2
+tensorboard==2.3.0
+torch
+torchvision
+Unidecode==1.1.1
+pyopenjtalk==0.2.0
+jamo==0.4.1
+pypinyin==0.44.0
+jieba==0.42.1
+cn2an==0.5.17
+jieba==0.42.1
+ipython==7.34.0
+gradio==3.4.1
+openai
+pydub
+inflect
+eng_to_ipa
+onnxruntime
\ No newline at end of file
diff --git a/text/LICENSE b/text/LICENSE
new file mode 100644
index 0000000000000000000000000000000000000000..4ad4ed1d5e34d95c8380768ec16405d789cc6de4
--- /dev/null
+++ b/text/LICENSE
@@ -0,0 +1,19 @@
+Copyright (c) 2017 Keith Ito
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
diff --git a/text/__init__.py b/text/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..48ae82f3e40ecd1bf17a7de78d87790327af3362
--- /dev/null
+++ b/text/__init__.py
@@ -0,0 +1,56 @@
+""" from https://github.com/keithito/tacotron """
+from text import cleaners
+from text.symbols import symbols
+
+
+# Mappings from symbol to numeric ID and vice versa:
+_symbol_to_id = {s: i for i, s in enumerate(symbols)}
+_id_to_symbol = {i: s for i, s in enumerate(symbols)}
+
+
+def text_to_sequence(text, cleaner_names):
+  '''Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+      cleaner_names: names of the cleaner functions to run the text through
+    Returns:
+      List of integers corresponding to the symbols in the text
+  '''
+  sequence = []
+
+  clean_text = _clean_text(text, cleaner_names)
+  for symbol in clean_text:
+    if symbol not in _symbol_to_id.keys():
+      continue
+    symbol_id = _symbol_to_id[symbol]
+    sequence += [symbol_id]
+  return sequence
+
+
+def cleaned_text_to_sequence(cleaned_text):
+  '''Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+    Returns:
+      List of integers corresponding to the symbols in the text
+  '''
+  sequence = [_symbol_to_id[symbol] for symbol in cleaned_text if symbol in _symbol_to_id.keys()]
+  return sequence
+
+
+def sequence_to_text(sequence):
+  '''Converts a sequence of IDs back to a string'''
+  result = ''
+  for symbol_id in sequence:
+    s = _id_to_symbol[symbol_id]
+    result += s
+  return result
+
+
+def _clean_text(text, cleaner_names):
+  for name in cleaner_names:
+    cleaner = getattr(cleaners, name)
+    if not cleaner:
+      raise Exception('Unknown cleaner: %s' % name)
+    text = cleaner(text)
+  return text
diff --git a/text/__pycache__/__init__.cpython-37.pyc b/text/__pycache__/__init__.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..8a791bce07f63f176c38217ad40b57d88ee9cf70
Binary files /dev/null and b/text/__pycache__/__init__.cpython-37.pyc differ
diff --git a/text/__pycache__/__init__.cpython-38.pyc b/text/__pycache__/__init__.cpython-38.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..73ec0b66e64e9ad68f5efc65689e4cd7ec7d5b41
Binary files /dev/null and b/text/__pycache__/__init__.cpython-38.pyc differ
diff --git a/text/__pycache__/__init__.cpython-39.pyc b/text/__pycache__/__init__.cpython-39.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..a21056f7c260b50726d61c2c1e0c35a1e77b29c7
Binary files /dev/null and b/text/__pycache__/__init__.cpython-39.pyc differ
diff --git a/text/__pycache__/cleaners.cpython-37.pyc b/text/__pycache__/cleaners.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..0211d13c1a049c10c856951d64e61e361dbc0a79
Binary files /dev/null and b/text/__pycache__/cleaners.cpython-37.pyc differ
diff --git a/text/__pycache__/cleaners.cpython-38.pyc b/text/__pycache__/cleaners.cpython-38.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..214b7862bd968990df4463fbee817aa3cf4a8602
Binary files /dev/null and b/text/__pycache__/cleaners.cpython-38.pyc differ
diff --git a/text/__pycache__/cleaners.cpython-39.pyc b/text/__pycache__/cleaners.cpython-39.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..a9f004ed1396da9b7be7235eddb3331faa9bad61
Binary files /dev/null and b/text/__pycache__/cleaners.cpython-39.pyc differ
diff --git a/text/__pycache__/english.cpython-37.pyc b/text/__pycache__/english.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..67af12d2984060609249e5b92b0ef96849f14db2
Binary files /dev/null and b/text/__pycache__/english.cpython-37.pyc differ
diff --git a/text/__pycache__/english.cpython-38.pyc b/text/__pycache__/english.cpython-38.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..6b3e6648a01b7ce179f2b56399705b0bf0b6e0f4
Binary files /dev/null and b/text/__pycache__/english.cpython-38.pyc differ
diff --git a/text/__pycache__/english.cpython-39.pyc b/text/__pycache__/english.cpython-39.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..2cc8a13bcb5ce9f9b2e85d322e848257576a23d8
Binary files /dev/null and b/text/__pycache__/english.cpython-39.pyc differ
diff --git a/text/__pycache__/japanese.cpython-37.pyc b/text/__pycache__/japanese.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..c5b21af0a9a15311da14c8cde5d7dad8dad6edfe
Binary files /dev/null and b/text/__pycache__/japanese.cpython-37.pyc differ
diff --git a/text/__pycache__/japanese.cpython-38.pyc b/text/__pycache__/japanese.cpython-38.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..bbdbb185b77ea81094492f63d1fcf9db62ca8fd2
Binary files /dev/null and b/text/__pycache__/japanese.cpython-38.pyc differ
diff --git a/text/__pycache__/japanese.cpython-39.pyc b/text/__pycache__/japanese.cpython-39.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..15b3c19c08a02b640aff770599e469171896f349
Binary files /dev/null and b/text/__pycache__/japanese.cpython-39.pyc differ
diff --git a/text/__pycache__/korean.cpython-37.pyc b/text/__pycache__/korean.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..cc7804b6ffdd617bbc312c02eab8bca33f7a49dc
Binary files /dev/null and b/text/__pycache__/korean.cpython-37.pyc differ
diff --git a/text/__pycache__/mandarin.cpython-37.pyc b/text/__pycache__/mandarin.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..bf0505a499ed9cd4fbef381b2bbb7c0c2ddb5b60
Binary files /dev/null and b/text/__pycache__/mandarin.cpython-37.pyc differ
diff --git a/text/__pycache__/mandarin.cpython-38.pyc b/text/__pycache__/mandarin.cpython-38.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..351cb45f946679dc1c124e568d995269a57e2154
Binary files /dev/null and b/text/__pycache__/mandarin.cpython-38.pyc differ
diff --git a/text/__pycache__/mandarin.cpython-39.pyc b/text/__pycache__/mandarin.cpython-39.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..5afc0dbfb99dac5ab34cdfe2a567ed22d322af9a
Binary files /dev/null and b/text/__pycache__/mandarin.cpython-39.pyc differ
diff --git a/text/__pycache__/sanskrit.cpython-37.pyc b/text/__pycache__/sanskrit.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..8fd202a3167dddb5dc444e904430ddddbda8fb21
Binary files /dev/null and b/text/__pycache__/sanskrit.cpython-37.pyc differ
diff --git a/text/__pycache__/shanghainese.cpython-37.pyc b/text/__pycache__/shanghainese.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..35832acc5a117ea38e3064bcb8827f5fd88c675e
Binary files /dev/null and b/text/__pycache__/shanghainese.cpython-37.pyc differ
diff --git a/text/__pycache__/symbols.cpython-37.pyc b/text/__pycache__/symbols.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..c3b5ef456a75a0b2a430f466d4cebac086966dc4
Binary files /dev/null and b/text/__pycache__/symbols.cpython-37.pyc differ
diff --git a/text/__pycache__/symbols.cpython-38.pyc b/text/__pycache__/symbols.cpython-38.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..8479e53ae2435f0adb421546b050730e2b90d1e8
Binary files /dev/null and b/text/__pycache__/symbols.cpython-38.pyc differ
diff --git a/text/__pycache__/symbols.cpython-39.pyc b/text/__pycache__/symbols.cpython-39.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..43cf34fa8459dcd073507ebbb0b3d9967708228e
Binary files /dev/null and b/text/__pycache__/symbols.cpython-39.pyc differ
diff --git a/text/__pycache__/thai.cpython-37.pyc b/text/__pycache__/thai.cpython-37.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..ba2b0d1fd908b42691c4d0a9a8494ad37a437c28
Binary files /dev/null and b/text/__pycache__/thai.cpython-37.pyc differ
diff --git a/text/cantonese.py b/text/cantonese.py
new file mode 100644
index 0000000000000000000000000000000000000000..b66d12138b81b70b86f18217d24a08fce76305c0
--- /dev/null
+++ b/text/cantonese.py
@@ -0,0 +1,59 @@
+import re
+import cn2an
+import opencc
+
+
+converter = opencc.OpenCC('jyutjyu')
+
+# List of (Latin alphabet, ipa) pairs:
+_latin_to_ipa = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('A', 'ei˥'),
+    ('B', 'biː˥'),
+    ('C', 'siː˥'),
+    ('D', 'tiː˥'),
+    ('E', 'iː˥'),
+    ('F', 'e˥fuː˨˩'),
+    ('G', 'tsiː˥'),
+    ('H', 'ɪk̚˥tsʰyː˨˩'),
+    ('I', 'ɐi˥'),
+    ('J', 'tsei˥'),
+    ('K', 'kʰei˥'),
+    ('L', 'e˥llou˨˩'),
+    ('M', 'ɛːm˥'),
+    ('N', 'ɛːn˥'),
+    ('O', 'ou˥'),
+    ('P', 'pʰiː˥'),
+    ('Q', 'kʰiːu˥'),
+    ('R', 'aː˥lou˨˩'),
+    ('S', 'ɛː˥siː˨˩'),
+    ('T', 'tʰiː˥'),
+    ('U', 'juː˥'),
+    ('V', 'wiː˥'),
+    ('W', 'tʊk̚˥piː˥juː˥'),
+    ('X', 'ɪk̚˥siː˨˩'),
+    ('Y', 'waːi˥'),
+    ('Z', 'iː˨sɛːt̚˥')
+]]
+
+
+def number_to_cantonese(text):
+    return re.sub(r'\d+(?:\.?\d+)?', lambda x: cn2an.an2cn(x.group()), text)
+
+
+def latin_to_ipa(text):
+    for regex, replacement in _latin_to_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def cantonese_to_ipa(text):
+    text = number_to_cantonese(text.upper())
+    text = converter.convert(text).replace('-','').replace('$',' ')
+    text = re.sub(r'[A-Z]', lambda x: latin_to_ipa(x.group())+' ', text)
+    text = re.sub(r'[、；：]', '，', text)
+    text = re.sub(r'\s*，\s*', ', ', text)
+    text = re.sub(r'\s*。\s*', '. ', text)
+    text = re.sub(r'\s*？\s*', '? ', text)
+    text = re.sub(r'\s*！\s*', '! ', text)
+    text = re.sub(r'\s*$', '', text)
+    return text
diff --git a/text/cleaners.py b/text/cleaners.py
new file mode 100644
index 0000000000000000000000000000000000000000..6405dd8a7b3b727b0d8a3cc418b50bd0d066d143
--- /dev/null
+++ b/text/cleaners.py
@@ -0,0 +1,176 @@
+import re
+from text.english import english_to_lazy_ipa, english_to_ipa2, english_to_lazy_ipa2
+from text.japanese import japanese_to_romaji_with_accent, japanese_to_ipa, japanese_to_ipa2, japanese_to_ipa3
+from text.mandarin import number_to_chinese, chinese_to_bopomofo, latin_to_bopomofo, chinese_to_romaji, chinese_to_lazy_ipa, chinese_to_ipa, chinese_to_ipa2
+# from text.sanskrit import devanagari_to_ipa
+# from text.english import english_to_lazy_ipa, english_to_ipa2, english_to_lazy_ipa2
+# from text.thai import num_to_thai, latin_to_thai
+# from text.shanghainese import shanghainese_to_ipa
+# from text.cantonese import cantonese_to_ipa
+# from text.ngu_dialect import ngu_dialect_to_ipa
+
+
+def japanese_cleaners(text):
+    text = japanese_to_romaji_with_accent(text)
+    if re.match('[A-Za-z]', text[-1]):
+        text += '.'
+    return text
+
+
+def japanese_cleaners2(text):
+    return japanese_cleaners(text).replace('ts', 'ʦ').replace('...', '…')
+
+
+def korean_cleaners(text):
+    '''Pipeline for Korean text'''
+    text = latin_to_hangul(text)
+    text = number_to_hangul(text)
+    text = divide_hangul(text)
+    if re.match('[\u3131-\u3163]', text[-1]):
+        text += '.'
+    return text
+
+
+def chinese_cleaners(text):
+    '''Pipeline for Chinese text'''
+    text = number_to_chinese(text)
+    text = chinese_to_bopomofo(text)
+    text = latin_to_bopomofo(text)
+    if re.match('[ˉˊˇˋ˙]', text[-1]):
+        text += '。'
+    return text
+
+
+def zh_ja_mixture_cleaners(text):
+    chinese_texts = re.findall(r'\[ZH\].*?\[ZH\]', text)
+    japanese_texts = re.findall(r'\[JA\].*?\[JA\]', text)
+    for chinese_text in chinese_texts:
+        cleaned_text = chinese_to_romaji(chinese_text[4:-4])
+        text = text.replace(chinese_text, cleaned_text+' ', 1)
+    for japanese_text in japanese_texts:
+        cleaned_text = japanese_to_romaji_with_accent(
+            japanese_text[4:-4]).replace('ts', 'ʦ').replace('u', 'ɯ').replace('...', '…')
+        text = text.replace(japanese_text, cleaned_text+' ', 1)
+    text = text[:-1]
+    if re.match('[A-Za-zɯɹəɥ→↓↑]', text[-1]):
+        text += '.'
+    return text
+
+
+def sanskrit_cleaners(text):
+    text = text.replace('॥', '।').replace('ॐ', 'ओम्')
+    if text[-1] != '।':
+        text += ' ।'
+    return text
+
+
+def cjks_cleaners(text):
+    chinese_texts = re.findall(r'\[ZH\].*?\[ZH\]', text)
+    japanese_texts = re.findall(r'\[JA\].*?\[JA\]', text)
+    korean_texts = re.findall(r'\[KO\].*?\[KO\]', text)
+    sanskrit_texts = re.findall(r'\[SA\].*?\[SA\]', text)
+    english_texts = re.findall(r'\[EN\].*?\[EN\]', text)
+    for chinese_text in chinese_texts:
+        cleaned_text = chinese_to_lazy_ipa(chinese_text[4:-4])
+        text = text.replace(chinese_text, cleaned_text+' ', 1)
+    for japanese_text in japanese_texts:
+        cleaned_text = japanese_to_ipa(japanese_text[4:-4])
+        text = text.replace(japanese_text, cleaned_text+' ', 1)
+    for korean_text in korean_texts:
+        cleaned_text = korean_to_lazy_ipa(korean_text[4:-4])
+        text = text.replace(korean_text, cleaned_text+' ', 1)
+    for sanskrit_text in sanskrit_texts:
+        cleaned_text = devanagari_to_ipa(sanskrit_text[4:-4])
+        text = text.replace(sanskrit_text, cleaned_text+' ', 1)
+    for english_text in english_texts:
+        cleaned_text = english_to_lazy_ipa(english_text[4:-4])
+        text = text.replace(english_text, cleaned_text+' ', 1)
+    text = text[:-1]
+    if re.match(r'[^\.,!\?\-…~]', text[-1]):
+        text += '.'
+    return text
+
+
+def cjke_cleaners(text):
+    chinese_texts = re.findall(r'\[ZH\].*?\[ZH\]', text)
+    japanese_texts = re.findall(r'\[JA\].*?\[JA\]', text)
+    korean_texts = re.findall(r'\[KO\].*?\[KO\]', text)
+    english_texts = re.findall(r'\[EN\].*?\[EN\]', text)
+    for chinese_text in chinese_texts:
+        cleaned_text = chinese_to_lazy_ipa(chinese_text[4:-4])
+        cleaned_text = cleaned_text.replace(
+            'ʧ', 'tʃ').replace('ʦ', 'ts').replace('ɥan', 'ɥæn')
+        text = text.replace(chinese_text, cleaned_text+' ', 1)
+    for japanese_text in japanese_texts:
+        cleaned_text = japanese_to_ipa(japanese_text[4:-4])
+        cleaned_text = cleaned_text.replace('ʧ', 'tʃ').replace(
+            'ʦ', 'ts').replace('ɥan', 'ɥæn').replace('ʥ', 'dz')
+        text = text.replace(japanese_text, cleaned_text+' ', 1)
+    for korean_text in korean_texts:
+        cleaned_text = korean_to_ipa(korean_text[4:-4])
+        text = text.replace(korean_text, cleaned_text+' ', 1)
+    for english_text in english_texts:
+        cleaned_text = english_to_ipa2(english_text[4:-4])
+        cleaned_text = cleaned_text.replace('ɑ', 'a').replace(
+            'ɔ', 'o').replace('ɛ', 'e').replace('ɪ', 'i').replace('ʊ', 'u')
+        text = text.replace(english_text, cleaned_text+' ', 1)
+    text = text[:-1]
+    if re.match(r'[^\.,!\?\-…~]', text[-1]):
+        text += '.'
+    return text
+
+
+def cjke_cleaners2(text):
+    chinese_texts = re.findall(r'\[ZH\].*?\[ZH\]', text)
+    japanese_texts = re.findall(r'\[JA\].*?\[JA\]', text)
+    korean_texts = re.findall(r'\[KO\].*?\[KO\]', text)
+    english_texts = re.findall(r'\[EN\].*?\[EN\]', text)
+    for chinese_text in chinese_texts:
+        cleaned_text = chinese_to_ipa(chinese_text[4:-4])
+        text = text.replace(chinese_text, cleaned_text+' ', 1)
+    for japanese_text in japanese_texts:
+        cleaned_text = japanese_to_ipa2(japanese_text[4:-4])
+        text = text.replace(japanese_text, cleaned_text+' ', 1)
+    for korean_text in korean_texts:
+        cleaned_text = korean_to_ipa(korean_text[4:-4])
+        text = text.replace(korean_text, cleaned_text+' ', 1)
+    for english_text in english_texts:
+        cleaned_text = english_to_ipa2(english_text[4:-4])
+        text = text.replace(english_text, cleaned_text+' ', 1)
+    text = text[:-1]
+    if re.match(r'[^\.,!\?\-…~]', text[-1]):
+        text += '.'
+    return text
+
+
+def thai_cleaners(text):
+    text = num_to_thai(text)
+    text = latin_to_thai(text)
+    return text
+
+
+def shanghainese_cleaners(text):
+    text = shanghainese_to_ipa(text)
+    if re.match(r'[^\.,!\?\-…~]', text[-1]):
+        text += '.'
+    return text
+
+
+def chinese_dialect_cleaners(text):
+    text = re.sub(r'\[MD\](.*?)\[MD\]',
+                  lambda x: chinese_to_ipa2(x.group(1))+' ', text)
+    text = re.sub(r'\[TW\](.*?)\[TW\]',
+                  lambda x: chinese_to_ipa2(x.group(1), True)+' ', text)
+    text = re.sub(r'\[JA\](.*?)\[JA\]',
+                  lambda x: japanese_to_ipa3(x.group(1)).replace('Q', 'ʔ')+' ', text)
+    text = re.sub(r'\[SH\](.*?)\[SH\]', lambda x: shanghainese_to_ipa(x.group(1)).replace('1', '˥˧').replace('5',
+                  '˧˧˦').replace('6', '˩˩˧').replace('7', '˥').replace('8', '˩˨').replace('ᴀ', 'ɐ').replace('ᴇ', 'e')+' ', text)
+    text = re.sub(r'\[GD\](.*?)\[GD\]',
+                  lambda x: cantonese_to_ipa(x.group(1))+' ', text)
+    text = re.sub(r'\[EN\](.*?)\[EN\]',
+                  lambda x: english_to_lazy_ipa2(x.group(1))+' ', text)
+    text = re.sub(r'\[([A-Z]{2})\](.*?)\[\1\]', lambda x: ngu_dialect_to_ipa(x.group(2), x.group(
+        1)).replace('ʣ', 'dz').replace('ʥ', 'dʑ').replace('ʦ', 'ts').replace('ʨ', 'tɕ')+' ', text)
+    text = re.sub(r'\s+$', '', text)
+    text = re.sub(r'([^\.,!\?\-…~])$', r'\1.', text)
+    return text
diff --git a/text/english.py b/text/english.py
new file mode 100644
index 0000000000000000000000000000000000000000..6817392ba8a9eb830351de89fb7afc5ad72f5e42
--- /dev/null
+++ b/text/english.py
@@ -0,0 +1,188 @@
+""" from https://github.com/keithito/tacotron """
+
+'''
+Cleaners are transformations that run over the input text at both training and eval time.
+
+Cleaners can be selected by passing a comma-delimited list of cleaner names as the "cleaners"
+hyperparameter. Some cleaners are English-specific. You'll typically want to use:
+  1. "english_cleaners" for English text
+  2. "transliteration_cleaners" for non-English text that can be transliterated to ASCII using
+     the Unidecode library (https://pypi.python.org/pypi/Unidecode)
+  3. "basic_cleaners" if you do not want to transliterate (in this case, you should also update
+     the symbols in symbols.py to match your data).
+'''
+
+
+# Regular expression matching whitespace:
+
+
+import re
+import inflect
+from unidecode import unidecode
+import eng_to_ipa as ipa
+_inflect = inflect.engine()
+_comma_number_re = re.compile(r'([0-9][0-9\,]+[0-9])')
+_decimal_number_re = re.compile(r'([0-9]+\.[0-9]+)')
+_pounds_re = re.compile(r'£([0-9\,]*[0-9]+)')
+_dollars_re = re.compile(r'\$([0-9\.\,]*[0-9]+)')
+_ordinal_re = re.compile(r'[0-9]+(st|nd|rd|th)')
+_number_re = re.compile(r'[0-9]+')
+
+# List of (regular expression, replacement) pairs for abbreviations:
+_abbreviations = [(re.compile('\\b%s\\.' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('mrs', 'misess'),
+    ('mr', 'mister'),
+    ('dr', 'doctor'),
+    ('st', 'saint'),
+    ('co', 'company'),
+    ('jr', 'junior'),
+    ('maj', 'major'),
+    ('gen', 'general'),
+    ('drs', 'doctors'),
+    ('rev', 'reverend'),
+    ('lt', 'lieutenant'),
+    ('hon', 'honorable'),
+    ('sgt', 'sergeant'),
+    ('capt', 'captain'),
+    ('esq', 'esquire'),
+    ('ltd', 'limited'),
+    ('col', 'colonel'),
+    ('ft', 'fort'),
+]]
+
+
+# List of (ipa, lazy ipa) pairs:
+_lazy_ipa = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('r', 'ɹ'),
+    ('æ', 'e'),
+    ('ɑ', 'a'),
+    ('ɔ', 'o'),
+    ('ð', 'z'),
+    ('θ', 's'),
+    ('ɛ', 'e'),
+    ('ɪ', 'i'),
+    ('ʊ', 'u'),
+    ('ʒ', 'ʥ'),
+    ('ʤ', 'ʥ'),
+    ('ˈ', '↓'),
+]]
+
+# List of (ipa, lazy ipa2) pairs:
+_lazy_ipa2 = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('r', 'ɹ'),
+    ('ð', 'z'),
+    ('θ', 's'),
+    ('ʒ', 'ʑ'),
+    ('ʤ', 'dʑ'),
+    ('ˈ', '↓'),
+]]
+
+# List of (ipa, ipa2) pairs
+_ipa_to_ipa2 = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('r', 'ɹ'),
+    ('ʤ', 'dʒ'),
+    ('ʧ', 'tʃ')
+]]
+
+
+def expand_abbreviations(text):
+    for regex, replacement in _abbreviations:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def collapse_whitespace(text):
+    return re.sub(r'\s+', ' ', text)
+
+
+def _remove_commas(m):
+    return m.group(1).replace(',', '')
+
+
+def _expand_decimal_point(m):
+    return m.group(1).replace('.', ' point ')
+
+
+def _expand_dollars(m):
+    match = m.group(1)
+    parts = match.split('.')
+    if len(parts) > 2:
+        return match + ' dollars'  # Unexpected format
+    dollars = int(parts[0]) if parts[0] else 0
+    cents = int(parts[1]) if len(parts) > 1 and parts[1] else 0
+    if dollars and cents:
+        dollar_unit = 'dollar' if dollars == 1 else 'dollars'
+        cent_unit = 'cent' if cents == 1 else 'cents'
+        return '%s %s, %s %s' % (dollars, dollar_unit, cents, cent_unit)
+    elif dollars:
+        dollar_unit = 'dollar' if dollars == 1 else 'dollars'
+        return '%s %s' % (dollars, dollar_unit)
+    elif cents:
+        cent_unit = 'cent' if cents == 1 else 'cents'
+        return '%s %s' % (cents, cent_unit)
+    else:
+        return 'zero dollars'
+
+
+def _expand_ordinal(m):
+    return _inflect.number_to_words(m.group(0))
+
+
+def _expand_number(m):
+    num = int(m.group(0))
+    if num > 1000 and num < 3000:
+        if num == 2000:
+            return 'two thousand'
+        elif num > 2000 and num < 2010:
+            return 'two thousand ' + _inflect.number_to_words(num % 100)
+        elif num % 100 == 0:
+            return _inflect.number_to_words(num // 100) + ' hundred'
+        else:
+            return _inflect.number_to_words(num, andword='', zero='oh', group=2).replace(', ', ' ')
+    else:
+        return _inflect.number_to_words(num, andword='')
+
+
+def normalize_numbers(text):
+    text = re.sub(_comma_number_re, _remove_commas, text)
+    text = re.sub(_pounds_re, r'\1 pounds', text)
+    text = re.sub(_dollars_re, _expand_dollars, text)
+    text = re.sub(_decimal_number_re, _expand_decimal_point, text)
+    text = re.sub(_ordinal_re, _expand_ordinal, text)
+    text = re.sub(_number_re, _expand_number, text)
+    return text
+
+
+def mark_dark_l(text):
+    return re.sub(r'l([^aeiouæɑɔəɛɪʊ ]*(?: |$))', lambda x: 'ɫ'+x.group(1), text)
+
+
+def english_to_ipa(text):
+    text = unidecode(text).lower()
+    text = expand_abbreviations(text)
+    text = normalize_numbers(text)
+    phonemes = ipa.convert(text)
+    phonemes = collapse_whitespace(phonemes)
+    return phonemes
+
+
+def english_to_lazy_ipa(text):
+    text = english_to_ipa(text)
+    for regex, replacement in _lazy_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def english_to_ipa2(text):
+    text = english_to_ipa(text)
+    text = mark_dark_l(text)
+    for regex, replacement in _ipa_to_ipa2:
+        text = re.sub(regex, replacement, text)
+    return text.replace('...', '…')
+
+
+def english_to_lazy_ipa2(text):
+    text = english_to_ipa(text)
+    for regex, replacement in _lazy_ipa2:
+        text = re.sub(regex, replacement, text)
+    return text
diff --git a/text/japanese.py b/text/japanese.py
new file mode 100644
index 0000000000000000000000000000000000000000..375e4d50872d5c68ee57ca17470a2ca425425eba
--- /dev/null
+++ b/text/japanese.py
@@ -0,0 +1,153 @@
+import re
+from unidecode import unidecode
+import pyopenjtalk
+
+
+# Regular expression matching Japanese without punctuation marks:
+_japanese_characters = re.compile(
+    r'[A-Za-z\d\u3005\u3040-\u30ff\u4e00-\u9fff\uff11-\uff19\uff21-\uff3a\uff41-\uff5a\uff66-\uff9d]')
+
+# Regular expression matching non-Japanese characters or punctuation marks:
+_japanese_marks = re.compile(
+    r'[^A-Za-z\d\u3005\u3040-\u30ff\u4e00-\u9fff\uff11-\uff19\uff21-\uff3a\uff41-\uff5a\uff66-\uff9d]')
+
+# List of (symbol, Japanese) pairs for marks:
+_symbols_to_japanese = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('％', 'パーセント')
+]]
+
+# List of (romaji, ipa) pairs for marks:
+_romaji_to_ipa = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('ts', 'ʦ'),
+    ('u', 'ɯ'),
+    ('j', 'ʥ'),
+    ('y', 'j'),
+    ('ni', 'n^i'),
+    ('nj', 'n^'),
+    ('hi', 'çi'),
+    ('hj', 'ç'),
+    ('f', 'ɸ'),
+    ('I', 'i*'),
+    ('U', 'ɯ*'),
+    ('r', 'ɾ')
+]]
+
+# List of (romaji, ipa2) pairs for marks:
+_romaji_to_ipa2 = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('u', 'ɯ'),
+    ('ʧ', 'tʃ'),
+    ('j', 'dʑ'),
+    ('y', 'j'),
+    ('ni', 'n^i'),
+    ('nj', 'n^'),
+    ('hi', 'çi'),
+    ('hj', 'ç'),
+    ('f', 'ɸ'),
+    ('I', 'i*'),
+    ('U', 'ɯ*'),
+    ('r', 'ɾ')
+]]
+
+# List of (consonant, sokuon) pairs:
+_real_sokuon = [(re.compile('%s' % x[0]), x[1]) for x in [
+    (r'Q([↑↓]*[kg])', r'k#\1'),
+    (r'Q([↑↓]*[tdjʧ])', r't#\1'),
+    (r'Q([↑↓]*[sʃ])', r's\1'),
+    (r'Q([↑↓]*[pb])', r'p#\1')
+]]
+
+# List of (consonant, hatsuon) pairs:
+_real_hatsuon = [(re.compile('%s' % x[0]), x[1]) for x in [
+    (r'N([↑↓]*[pbm])', r'm\1'),
+    (r'N([↑↓]*[ʧʥj])', r'n^\1'),
+    (r'N([↑↓]*[tdn])', r'n\1'),
+    (r'N([↑↓]*[kg])', r'ŋ\1')
+]]
+
+
+def symbols_to_japanese(text):
+    for regex, replacement in _symbols_to_japanese:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def japanese_to_romaji_with_accent(text):
+    '''Reference https://r9y9.github.io/ttslearn/latest/notebooks/ch10_Recipe-Tacotron.html'''
+    text = symbols_to_japanese(text)
+    sentences = re.split(_japanese_marks, text)
+    marks = re.findall(_japanese_marks, text)
+    text = ''
+    for i, sentence in enumerate(sentences):
+        if re.match(_japanese_characters, sentence):
+            if text != '':
+                text += ' '
+            labels = pyopenjtalk.extract_fullcontext(sentence)
+            for n, label in enumerate(labels):
+                phoneme = re.search(r'\-([^\+]*)\+', label).group(1)
+                if phoneme not in ['sil', 'pau']:
+                    text += phoneme.replace('ch', 'ʧ').replace('sh',
+                                                               'ʃ').replace('cl', 'Q')
+                else:
+                    continue
+                # n_moras = int(re.search(r'/F:(\d+)_', label).group(1))
+                a1 = int(re.search(r"/A:(\-?[0-9]+)\+", label).group(1))
+                a2 = int(re.search(r"\+(\d+)\+", label).group(1))
+                a3 = int(re.search(r"\+(\d+)/", label).group(1))
+                if re.search(r'\-([^\+]*)\+', labels[n + 1]).group(1) in ['sil', 'pau']:
+                    a2_next = -1
+                else:
+                    a2_next = int(
+                        re.search(r"\+(\d+)\+", labels[n + 1]).group(1))
+                # Accent phrase boundary
+                if a3 == 1 and a2_next == 1:
+                    text += ' '
+                # Falling
+                elif a1 == 0 and a2_next == a2 + 1:
+                    text += '↓'
+                # Rising
+                elif a2 == 1 and a2_next == 2:
+                    text += '↑'
+        if i < len(marks):
+            text += unidecode(marks[i]).replace(' ', '')
+    return text
+
+
+def get_real_sokuon(text):
+    for regex, replacement in _real_sokuon:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def get_real_hatsuon(text):
+    for regex, replacement in _real_hatsuon:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def japanese_to_ipa(text):
+    text = japanese_to_romaji_with_accent(text).replace('...', '…')
+    text = re.sub(
+        r'([aiueo])\1+', lambda x: x.group(0)[0]+'ː'*(len(x.group(0))-1), text)
+    text = get_real_sokuon(text)
+    text = get_real_hatsuon(text)
+    for regex, replacement in _romaji_to_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def japanese_to_ipa2(text):
+    text = japanese_to_romaji_with_accent(text).replace('...', '…')
+    text = get_real_sokuon(text)
+    text = get_real_hatsuon(text)
+    for regex, replacement in _romaji_to_ipa2:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def japanese_to_ipa3(text):
+    text = japanese_to_ipa2(text).replace('n^', 'ȵ').replace(
+        'ʃ', 'ɕ').replace('*', '\u0325').replace('#', '\u031a')
+    text = re.sub(
+        r'([aiɯeo])\1+', lambda x: x.group(0)[0]+'ː'*(len(x.group(0))-1), text)
+    text = re.sub(r'((?:^|\s)(?:ts|tɕ|[kpt]))', r'\1ʰ', text)
+    return text
diff --git a/text/korean.py b/text/korean.py
new file mode 100644
index 0000000000000000000000000000000000000000..edee07429a450c55e3d8e246997faaa1e0b89cc9
--- /dev/null
+++ b/text/korean.py
@@ -0,0 +1,210 @@
+import re
+from jamo import h2j, j2hcj
+import ko_pron
+
+
+# This is a list of Korean classifiers preceded by pure Korean numerals.
+_korean_classifiers = '군데 권 개 그루 닢 대 두 마리 모 모금 뭇 발 발짝 방 번 벌 보루 살 수 술 시 쌈 움큼 정 짝 채 척 첩 축 켤레 톨 통'
+
+# List of (hangul, hangul divided) pairs:
+_hangul_divided = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('ㄳ', 'ㄱㅅ'),
+    ('ㄵ', 'ㄴㅈ'),
+    ('ㄶ', 'ㄴㅎ'),
+    ('ㄺ', 'ㄹㄱ'),
+    ('ㄻ', 'ㄹㅁ'),
+    ('ㄼ', 'ㄹㅂ'),
+    ('ㄽ', 'ㄹㅅ'),
+    ('ㄾ', 'ㄹㅌ'),
+    ('ㄿ', 'ㄹㅍ'),
+    ('ㅀ', 'ㄹㅎ'),
+    ('ㅄ', 'ㅂㅅ'),
+    ('ㅘ', 'ㅗㅏ'),
+    ('ㅙ', 'ㅗㅐ'),
+    ('ㅚ', 'ㅗㅣ'),
+    ('ㅝ', 'ㅜㅓ'),
+    ('ㅞ', 'ㅜㅔ'),
+    ('ㅟ', 'ㅜㅣ'),
+    ('ㅢ', 'ㅡㅣ'),
+    ('ㅑ', 'ㅣㅏ'),
+    ('ㅒ', 'ㅣㅐ'),
+    ('ㅕ', 'ㅣㅓ'),
+    ('ㅖ', 'ㅣㅔ'),
+    ('ㅛ', 'ㅣㅗ'),
+    ('ㅠ', 'ㅣㅜ')
+]]
+
+# List of (Latin alphabet, hangul) pairs:
+_latin_to_hangul = [(re.compile('%s' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('a', '에이'),
+    ('b', '비'),
+    ('c', '시'),
+    ('d', '디'),
+    ('e', '이'),
+    ('f', '에프'),
+    ('g', '지'),
+    ('h', '에이치'),
+    ('i', '아이'),
+    ('j', '제이'),
+    ('k', '케이'),
+    ('l', '엘'),
+    ('m', '엠'),
+    ('n', '엔'),
+    ('o', '오'),
+    ('p', '피'),
+    ('q', '큐'),
+    ('r', '아르'),
+    ('s', '에스'),
+    ('t', '티'),
+    ('u', '유'),
+    ('v', '브이'),
+    ('w', '더블유'),
+    ('x', '엑스'),
+    ('y', '와이'),
+    ('z', '제트')
+]]
+
+# List of (ipa, lazy ipa) pairs:
+_ipa_to_lazy_ipa = [(re.compile('%s' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('t͡ɕ','ʧ'),
+    ('d͡ʑ','ʥ'),
+    ('ɲ','n^'),
+    ('ɕ','ʃ'),
+    ('ʷ','w'),
+    ('ɭ','l`'),
+    ('ʎ','ɾ'),
+    ('ɣ','ŋ'),
+    ('ɰ','ɯ'),
+    ('ʝ','j'),
+    ('ʌ','ə'),
+    ('ɡ','g'),
+    ('\u031a','#'),
+    ('\u0348','='),
+    ('\u031e',''),
+    ('\u0320',''),
+    ('\u0339','')
+]]
+
+
+def latin_to_hangul(text):
+    for regex, replacement in _latin_to_hangul:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def divide_hangul(text):
+    text = j2hcj(h2j(text))
+    for regex, replacement in _hangul_divided:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def hangul_number(num, sino=True):
+    '''Reference https://github.com/Kyubyong/g2pK'''
+    num = re.sub(',', '', num)
+
+    if num == '0':
+        return '영'
+    if not sino and num == '20':
+        return '스무'
+
+    digits = '123456789'
+    names = '일이삼사오육칠팔구'
+    digit2name = {d: n for d, n in zip(digits, names)}
+
+    modifiers = '한 두 세 네 다섯 여섯 일곱 여덟 아홉'
+    decimals = '열 스물 서른 마흔 쉰 예순 일흔 여든 아흔'
+    digit2mod = {d: mod for d, mod in zip(digits, modifiers.split())}
+    digit2dec = {d: dec for d, dec in zip(digits, decimals.split())}
+
+    spelledout = []
+    for i, digit in enumerate(num):
+        i = len(num) - i - 1
+        if sino:
+            if i == 0:
+                name = digit2name.get(digit, '')
+            elif i == 1:
+                name = digit2name.get(digit, '') + '십'
+                name = name.replace('일십', '십')
+        else:
+            if i == 0:
+                name = digit2mod.get(digit, '')
+            elif i == 1:
+                name = digit2dec.get(digit, '')
+        if digit == '0':
+            if i % 4 == 0:
+                last_three = spelledout[-min(3, len(spelledout)):]
+                if ''.join(last_three) == '':
+                    spelledout.append('')
+                    continue
+            else:
+                spelledout.append('')
+                continue
+        if i == 2:
+            name = digit2name.get(digit, '') + '백'
+            name = name.replace('일백', '백')
+        elif i == 3:
+            name = digit2name.get(digit, '') + '천'
+            name = name.replace('일천', '천')
+        elif i == 4:
+            name = digit2name.get(digit, '') + '만'
+            name = name.replace('일만', '만')
+        elif i == 5:
+            name = digit2name.get(digit, '') + '십'
+            name = name.replace('일십', '십')
+        elif i == 6:
+            name = digit2name.get(digit, '') + '백'
+            name = name.replace('일백', '백')
+        elif i == 7:
+            name = digit2name.get(digit, '') + '천'
+            name = name.replace('일천', '천')
+        elif i == 8:
+            name = digit2name.get(digit, '') + '억'
+        elif i == 9:
+            name = digit2name.get(digit, '') + '십'
+        elif i == 10:
+            name = digit2name.get(digit, '') + '백'
+        elif i == 11:
+            name = digit2name.get(digit, '') + '천'
+        elif i == 12:
+            name = digit2name.get(digit, '') + '조'
+        elif i == 13:
+            name = digit2name.get(digit, '') + '십'
+        elif i == 14:
+            name = digit2name.get(digit, '') + '백'
+        elif i == 15:
+            name = digit2name.get(digit, '') + '천'
+        spelledout.append(name)
+    return ''.join(elem for elem in spelledout)
+
+
+def number_to_hangul(text):
+    '''Reference https://github.com/Kyubyong/g2pK'''
+    tokens = set(re.findall(r'(\d[\d,]*)([\uac00-\ud71f]+)', text))
+    for token in tokens:
+        num, classifier = token
+        if classifier[:2] in _korean_classifiers or classifier[0] in _korean_classifiers:
+            spelledout = hangul_number(num, sino=False)
+        else:
+            spelledout = hangul_number(num, sino=True)
+        text = text.replace(f'{num}{classifier}', f'{spelledout}{classifier}')
+    # digit by digit for remaining digits
+    digits = '0123456789'
+    names = '영일이삼사오육칠팔구'
+    for d, n in zip(digits, names):
+        text = text.replace(d, n)
+    return text
+
+
+def korean_to_lazy_ipa(text):
+    text = latin_to_hangul(text)
+    text = number_to_hangul(text)
+    text=re.sub('[\uac00-\ud7af]+',lambda x:ko_pron.romanise(x.group(0),'ipa').split('] ~ [')[0],text)
+    for regex, replacement in _ipa_to_lazy_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def korean_to_ipa(text):
+    text = korean_to_lazy_ipa(text)
+    return text.replace('ʧ','tʃ').replace('ʥ','dʑ')
diff --git a/text/mandarin.py b/text/mandarin.py
new file mode 100644
index 0000000000000000000000000000000000000000..a9ce0c4b223cd7fbb00e8332d2dd53de4c7cea09
--- /dev/null
+++ b/text/mandarin.py
@@ -0,0 +1,328 @@
+import os
+import sys
+import re
+from pypinyin import lazy_pinyin, BOPOMOFO
+import jieba
+import cn2an
+
+
+# List of (Latin alphabet, bopomofo) pairs:
+_latin_to_bopomofo = [(re.compile('%s' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('a', 'ㄟˉ'),
+    ('b', 'ㄅㄧˋ'),
+    ('c', 'ㄙㄧˉ'),
+    ('d', 'ㄉㄧˋ'),
+    ('e', 'ㄧˋ'),
+    ('f', 'ㄝˊㄈㄨˋ'),
+    ('g', 'ㄐㄧˋ'),
+    ('h', 'ㄝˇㄑㄩˋ'),
+    ('i', 'ㄞˋ'),
+    ('j', 'ㄐㄟˋ'),
+    ('k', 'ㄎㄟˋ'),
+    ('l', 'ㄝˊㄛˋ'),
+    ('m', 'ㄝˊㄇㄨˋ'),
+    ('n', 'ㄣˉ'),
+    ('o', 'ㄡˉ'),
+    ('p', 'ㄆㄧˉ'),
+    ('q', 'ㄎㄧㄡˉ'),
+    ('r', 'ㄚˋ'),
+    ('s', 'ㄝˊㄙˋ'),
+    ('t', 'ㄊㄧˋ'),
+    ('u', 'ㄧㄡˉ'),
+    ('v', 'ㄨㄧˉ'),
+    ('w', 'ㄉㄚˋㄅㄨˋㄌㄧㄡˋ'),
+    ('x', 'ㄝˉㄎㄨˋㄙˋ'),
+    ('y', 'ㄨㄞˋ'),
+    ('z', 'ㄗㄟˋ')
+]]
+
+# List of (bopomofo, romaji) pairs:
+_bopomofo_to_romaji = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('ㄅㄛ', 'p⁼wo'),
+    ('ㄆㄛ', 'pʰwo'),
+    ('ㄇㄛ', 'mwo'),
+    ('ㄈㄛ', 'fwo'),
+    ('ㄅ', 'p⁼'),
+    ('ㄆ', 'pʰ'),
+    ('ㄇ', 'm'),
+    ('ㄈ', 'f'),
+    ('ㄉ', 't⁼'),
+    ('ㄊ', 'tʰ'),
+    ('ㄋ', 'n'),
+    ('ㄌ', 'l'),
+    ('ㄍ', 'k⁼'),
+    ('ㄎ', 'kʰ'),
+    ('ㄏ', 'h'),
+    ('ㄐ', 'ʧ⁼'),
+    ('ㄑ', 'ʧʰ'),
+    ('ㄒ', 'ʃ'),
+    ('ㄓ', 'ʦ`⁼'),
+    ('ㄔ', 'ʦ`ʰ'),
+    ('ㄕ', 's`'),
+    ('ㄖ', 'ɹ`'),
+    ('ㄗ', 'ʦ⁼'),
+    ('ㄘ', 'ʦʰ'),
+    ('ㄙ', 's'),
+    ('ㄚ', 'a'),
+    ('ㄛ', 'o'),
+    ('ㄜ', 'ə'),
+    ('ㄝ', 'e'),
+    ('ㄞ', 'ai'),
+    ('ㄟ', 'ei'),
+    ('ㄠ', 'au'),
+    ('ㄡ', 'ou'),
+    ('ㄧㄢ', 'yeNN'),
+    ('ㄢ', 'aNN'),
+    ('ㄧㄣ', 'iNN'),
+    ('ㄣ', 'əNN'),
+    ('ㄤ', 'aNg'),
+    ('ㄧㄥ', 'iNg'),
+    ('ㄨㄥ', 'uNg'),
+    ('ㄩㄥ', 'yuNg'),
+    ('ㄥ', 'əNg'),
+    ('ㄦ', 'əɻ'),
+    ('ㄧ', 'i'),
+    ('ㄨ', 'u'),
+    ('ㄩ', 'ɥ'),
+    ('ˉ', '→'),
+    ('ˊ', '↑'),
+    ('ˇ', '↓↑'),
+    ('ˋ', '↓'),
+    ('˙', ''),
+    ('，', ','),
+    ('。', '.'),
+    ('！', '!'),
+    ('？', '?'),
+    ('—', '-')
+]]
+
+# List of (romaji, ipa) pairs:
+_romaji_to_ipa = [(re.compile('%s' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('ʃy', 'ʃ'),
+    ('ʧʰy', 'ʧʰ'),
+    ('ʧ⁼y', 'ʧ⁼'),
+    ('NN', 'n'),
+    ('Ng', 'ŋ'),
+    ('y', 'j'),
+    ('h', 'x')
+]]
+
+# List of (bopomofo, ipa) pairs:
+_bopomofo_to_ipa = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('ㄅㄛ', 'p⁼wo'),
+    ('ㄆㄛ', 'pʰwo'),
+    ('ㄇㄛ', 'mwo'),
+    ('ㄈㄛ', 'fwo'),
+    ('ㄅ', 'p⁼'),
+    ('ㄆ', 'pʰ'),
+    ('ㄇ', 'm'),
+    ('ㄈ', 'f'),
+    ('ㄉ', 't⁼'),
+    ('ㄊ', 'tʰ'),
+    ('ㄋ', 'n'),
+    ('ㄌ', 'l'),
+    ('ㄍ', 'k⁼'),
+    ('ㄎ', 'kʰ'),
+    ('ㄏ', 'x'),
+    ('ㄐ', 'tʃ⁼'),
+    ('ㄑ', 'tʃʰ'),
+    ('ㄒ', 'ʃ'),
+    ('ㄓ', 'ts`⁼'),
+    ('ㄔ', 'ts`ʰ'),
+    ('ㄕ', 's`'),
+    ('ㄖ', 'ɹ`'),
+    ('ㄗ', 'ts⁼'),
+    ('ㄘ', 'tsʰ'),
+    ('ㄙ', 's'),
+    ('ㄚ', 'a'),
+    ('ㄛ', 'o'),
+    ('ㄜ', 'ə'),
+    ('ㄝ', 'ɛ'),
+    ('ㄞ', 'aɪ'),
+    ('ㄟ', 'eɪ'),
+    ('ㄠ', 'ɑʊ'),
+    ('ㄡ', 'oʊ'),
+    ('ㄧㄢ', 'jɛn'),
+    ('ㄩㄢ', 'ɥæn'),
+    ('ㄢ', 'an'),
+    ('ㄧㄣ', 'in'),
+    ('ㄩㄣ', 'ɥn'),
+    ('ㄣ', 'ən'),
+    ('ㄤ', 'ɑŋ'),
+    ('ㄧㄥ', 'iŋ'),
+    ('ㄨㄥ', 'ʊŋ'),
+    ('ㄩㄥ', 'jʊŋ'),
+    ('ㄥ', 'əŋ'),
+    ('ㄦ', 'əɻ'),
+    ('ㄧ', 'i'),
+    ('ㄨ', 'u'),
+    ('ㄩ', 'ɥ'),
+    ('ˉ', '→'),
+    ('ˊ', '↑'),
+    ('ˇ', '↓↑'),
+    ('ˋ', '↓'),
+    ('˙', ''),
+    ('，', ','),
+    ('。', '.'),
+    ('！', '!'),
+    ('？', '?'),
+    ('—', '-')
+]]
+
+# List of (bopomofo, ipa2) pairs:
+_bopomofo_to_ipa2 = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('ㄅㄛ', 'pwo'),
+    ('ㄆㄛ', 'pʰwo'),
+    ('ㄇㄛ', 'mwo'),
+    ('ㄈㄛ', 'fwo'),
+    ('ㄅ', 'p'),
+    ('ㄆ', 'pʰ'),
+    ('ㄇ', 'm'),
+    ('ㄈ', 'f'),
+    ('ㄉ', 't'),
+    ('ㄊ', 'tʰ'),
+    ('ㄋ', 'n'),
+    ('ㄌ', 'l'),
+    ('ㄍ', 'k'),
+    ('ㄎ', 'kʰ'),
+    ('ㄏ', 'h'),
+    ('ㄐ', 'tɕ'),
+    ('ㄑ', 'tɕʰ'),
+    ('ㄒ', 'ɕ'),
+    ('ㄓ', 'tʂ'),
+    ('ㄔ', 'tʂʰ'),
+    ('ㄕ', 'ʂ'),
+    ('ㄖ', 'ɻ'),
+    ('ㄗ', 'ts'),
+    ('ㄘ', 'tsʰ'),
+    ('ㄙ', 's'),
+    ('ㄚ', 'a'),
+    ('ㄛ', 'o'),
+    ('ㄜ', 'ɤ'),
+    ('ㄝ', 'ɛ'),
+    ('ㄞ', 'aɪ'),
+    ('ㄟ', 'eɪ'),
+    ('ㄠ', 'ɑʊ'),
+    ('ㄡ', 'oʊ'),
+    ('ㄧㄢ', 'jɛn'),
+    ('ㄩㄢ', 'yæn'),
+    ('ㄢ', 'an'),
+    ('ㄧㄣ', 'in'),
+    ('ㄩㄣ', 'yn'),
+    ('ㄣ', 'ən'),
+    ('ㄤ', 'ɑŋ'),
+    ('ㄧㄥ', 'iŋ'),
+    ('ㄨㄥ', 'ʊŋ'),
+    ('ㄩㄥ', 'jʊŋ'),
+    ('ㄥ', 'ɤŋ'),
+    ('ㄦ', 'əɻ'),
+    ('ㄧ', 'i'),
+    ('ㄨ', 'u'),
+    ('ㄩ', 'y'),
+    ('ˉ', '˥'),
+    ('ˊ', '˧˥'),
+    ('ˇ', '˨˩˦'),
+    ('ˋ', '˥˩'),
+    ('˙', ''),
+    ('，', ','),
+    ('。', '.'),
+    ('！', '!'),
+    ('？', '?'),
+    ('—', '-')
+]]
+
+
+def number_to_chinese(text):
+    numbers = re.findall(r'\d+(?:\.?\d+)?', text)
+    for number in numbers:
+        text = text.replace(number, cn2an.an2cn(number), 1)
+    return text
+
+
+def chinese_to_bopomofo(text, taiwanese=False):
+    text = text.replace('、', '，').replace('；', '，').replace('：', '，')
+    words = jieba.lcut(text, cut_all=False)
+    text = ''
+    for word in words:
+        bopomofos = lazy_pinyin(word, BOPOMOFO)
+        if not re.search('[\u4e00-\u9fff]', word):
+            text += word
+            continue
+        for i in range(len(bopomofos)):
+            bopomofos[i] = re.sub(r'([\u3105-\u3129])$', r'\1ˉ', bopomofos[i])
+        if text != '':
+            text += ' '
+        if taiwanese:
+            text += '#'+'#'.join(bopomofos)
+        else:
+            text += ''.join(bopomofos)
+    return text
+
+
+def latin_to_bopomofo(text):
+    for regex, replacement in _latin_to_bopomofo:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def bopomofo_to_romaji(text):
+    for regex, replacement in _bopomofo_to_romaji:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def bopomofo_to_ipa(text):
+    for regex, replacement in _bopomofo_to_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def bopomofo_to_ipa2(text):
+    for regex, replacement in _bopomofo_to_ipa2:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def chinese_to_romaji(text):
+    text = number_to_chinese(text)
+    text = chinese_to_bopomofo(text)
+    text = latin_to_bopomofo(text)
+    text = bopomofo_to_romaji(text)
+    text = re.sub('i([aoe])', r'y\1', text)
+    text = re.sub('u([aoəe])', r'w\1', text)
+    text = re.sub('([ʦsɹ]`[⁼ʰ]?)([→↓↑ ]+|$)',
+                  r'\1ɹ`\2', text).replace('ɻ', 'ɹ`')
+    text = re.sub('([ʦs][⁼ʰ]?)([→↓↑ ]+|$)', r'\1ɹ\2', text)
+    return text
+
+
+def chinese_to_lazy_ipa(text):
+    text = chinese_to_romaji(text)
+    for regex, replacement in _romaji_to_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def chinese_to_ipa(text):
+    text = number_to_chinese(text)
+    text = chinese_to_bopomofo(text)
+    text = latin_to_bopomofo(text)
+    text = bopomofo_to_ipa(text)
+    text = re.sub('i([aoe])', r'j\1', text)
+    text = re.sub('u([aoəe])', r'w\1', text)
+    text = re.sub('([sɹ]`[⁼ʰ]?)([→↓↑ ]+|$)',
+                  r'\1ɹ`\2', text).replace('ɻ', 'ɹ`')
+    text = re.sub('([s][⁼ʰ]?)([→↓↑ ]+|$)', r'\1ɹ\2', text)
+    return text
+
+
+def chinese_to_ipa2(text, taiwanese=False):
+    text = number_to_chinese(text)
+    text = chinese_to_bopomofo(text, taiwanese)
+    text = latin_to_bopomofo(text)
+    text = bopomofo_to_ipa2(text)
+    text = re.sub(r'i([aoe])', r'j\1', text)
+    text = re.sub(r'u([aoəe])', r'w\1', text)
+    text = re.sub(r'([ʂɹ]ʰ?)([˩˨˧˦˥ ]+|$)', r'\1ʅ\2', text)
+    text = re.sub(r'(sʰ?)([˩˨˧˦˥ ]+|$)', r'\1ɿ\2', text)
+    return text
diff --git a/text/ngu_dialect.py b/text/ngu_dialect.py
new file mode 100644
index 0000000000000000000000000000000000000000..ce3e12bbf0469426872eed5f681985d3e1be9b26
--- /dev/null
+++ b/text/ngu_dialect.py
@@ -0,0 +1,30 @@
+import re
+import opencc
+
+
+dialects = {'SZ': 'suzhou', 'WX': 'wuxi', 'CZ': 'changzhou', 'HZ': 'hangzhou',
+            'SX': 'shaoxing', 'NB': 'ningbo', 'JJ': 'jingjiang', 'YX': 'yixing',
+            'JD': 'jiading', 'ZR': 'zhenru', 'PH': 'pinghu', 'TX': 'tongxiang',
+            'JS': 'jiashan', 'HN': 'xiashi', 'LP': 'linping', 'XS': 'xiaoshan',
+            'FY': 'fuyang', 'RA': 'ruao', 'CX': 'cixi', 'SM': 'sanmen',
+            'TT': 'tiantai', 'WZ': 'wenzhou', 'SC': 'suichang', 'YB': 'youbu'}
+
+converters = {}
+
+for dialect in dialects.values():
+    try:
+        converters[dialect] = opencc.OpenCC(dialect)
+    except:
+        pass
+
+
+def ngu_dialect_to_ipa(text, dialect):
+    dialect = dialects[dialect]
+    text = converters[dialect].convert(text).replace('-','').replace('$',' ')
+    text = re.sub(r'[、；：]', '，', text)
+    text = re.sub(r'\s*，\s*', ', ', text)
+    text = re.sub(r'\s*。\s*', '. ', text)
+    text = re.sub(r'\s*？\s*', '? ', text)
+    text = re.sub(r'\s*！\s*', '! ', text)
+    text = re.sub(r'\s*$', '', text)
+    return text
diff --git a/text/sanskrit.py b/text/sanskrit.py
new file mode 100644
index 0000000000000000000000000000000000000000..0223aaac384a2f850f5bc20651fc18eb964607d0
--- /dev/null
+++ b/text/sanskrit.py
@@ -0,0 +1,62 @@
+import re
+from indic_transliteration import sanscript
+
+
+# List of (iast, ipa) pairs:
+_iast_to_ipa = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('a', 'ə'),
+    ('ā', 'aː'),
+    ('ī', 'iː'),
+    ('ū', 'uː'),
+    ('ṛ', 'ɹ`'),
+    ('ṝ', 'ɹ`ː'),
+    ('ḷ', 'l`'),
+    ('ḹ', 'l`ː'),
+    ('e', 'eː'),
+    ('o', 'oː'),
+    ('k', 'k⁼'),
+    ('k⁼h', 'kʰ'),
+    ('g', 'g⁼'),
+    ('g⁼h', 'gʰ'),
+    ('ṅ', 'ŋ'),
+    ('c', 'ʧ⁼'),
+    ('ʧ⁼h', 'ʧʰ'),
+    ('j', 'ʥ⁼'),
+    ('ʥ⁼h', 'ʥʰ'),
+    ('ñ', 'n^'),
+    ('ṭ', 't`⁼'),
+    ('t`⁼h', 't`ʰ'),
+    ('ḍ', 'd`⁼'),
+    ('d`⁼h', 'd`ʰ'),
+    ('ṇ', 'n`'),
+    ('t', 't⁼'),
+    ('t⁼h', 'tʰ'),
+    ('d', 'd⁼'),
+    ('d⁼h', 'dʰ'),
+    ('p', 'p⁼'),
+    ('p⁼h', 'pʰ'),
+    ('b', 'b⁼'),
+    ('b⁼h', 'bʰ'),
+    ('y', 'j'),
+    ('ś', 'ʃ'),
+    ('ṣ', 's`'),
+    ('r', 'ɾ'),
+    ('l̤', 'l`'),
+    ('h', 'ɦ'),
+    ("'", ''),
+    ('~', '^'),
+    ('ṃ', '^')
+]]
+
+
+def devanagari_to_ipa(text):
+    text = text.replace('ॐ', 'ओम्')
+    text = re.sub(r'\s*।\s*$', '.', text)
+    text = re.sub(r'\s*।\s*', ', ', text)
+    text = re.sub(r'\s*॥', '.', text)
+    text = sanscript.transliterate(text, sanscript.DEVANAGARI, sanscript.IAST)
+    for regex, replacement in _iast_to_ipa:
+        text = re.sub(regex, replacement, text)
+    text = re.sub('(.)[`ː]*ḥ', lambda x: x.group(0)
+                  [:-1]+'h'+x.group(1)+'*', text)
+    return text
diff --git a/text/shanghainese.py b/text/shanghainese.py
new file mode 100644
index 0000000000000000000000000000000000000000..cb29c24a08d2e406e8399cf7bc9fe5cb43cb9c61
--- /dev/null
+++ b/text/shanghainese.py
@@ -0,0 +1,64 @@
+import re
+import cn2an
+import opencc
+
+
+converter = opencc.OpenCC('zaonhe')
+
+# List of (Latin alphabet, ipa) pairs:
+_latin_to_ipa = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('A', 'ᴇ'),
+    ('B', 'bi'),
+    ('C', 'si'),
+    ('D', 'di'),
+    ('E', 'i'),
+    ('F', 'ᴇf'),
+    ('G', 'dʑi'),
+    ('H', 'ᴇtɕʰ'),
+    ('I', 'ᴀi'),
+    ('J', 'dʑᴇ'),
+    ('K', 'kʰᴇ'),
+    ('L', 'ᴇl'),
+    ('M', 'ᴇm'),
+    ('N', 'ᴇn'),
+    ('O', 'o'),
+    ('P', 'pʰi'),
+    ('Q', 'kʰiu'),
+    ('R', 'ᴀl'),
+    ('S', 'ᴇs'),
+    ('T', 'tʰi'),
+    ('U', 'ɦiu'),
+    ('V', 'vi'),
+    ('W', 'dᴀbɤliu'),
+    ('X', 'ᴇks'),
+    ('Y', 'uᴀi'),
+    ('Z', 'zᴇ')
+]]
+
+
+def _number_to_shanghainese(num):
+    num = cn2an.an2cn(num).replace('一十','十').replace('二十', '廿').replace('二', '两')
+    return re.sub(r'((?:^|[^三四五六七八九])十|廿)两', r'\1二', num)
+
+
+def number_to_shanghainese(text):
+    return re.sub(r'\d+(?:\.?\d+)?', lambda x: _number_to_shanghainese(x.group()), text)
+
+
+def latin_to_ipa(text):
+    for regex, replacement in _latin_to_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def shanghainese_to_ipa(text):
+    text = number_to_shanghainese(text.upper())
+    text = converter.convert(text).replace('-','').replace('$',' ')
+    text = re.sub(r'[A-Z]', lambda x: latin_to_ipa(x.group())+' ', text)
+    text = re.sub(r'[、；：]', '，', text)
+    text = re.sub(r'\s*，\s*', ', ', text)
+    text = re.sub(r'\s*。\s*', '. ', text)
+    text = re.sub(r'\s*？\s*', '? ', text)
+    text = re.sub(r'\s*！\s*', '! ', text)
+    text = re.sub(r'\s*$', '', text)
+    return text
diff --git a/text/symbols.py b/text/symbols.py
new file mode 100644
index 0000000000000000000000000000000000000000..3705de1c96d52d5643eab9bc80671fe9cb7e4363
--- /dev/null
+++ b/text/symbols.py
@@ -0,0 +1,67 @@
+'''
+Defines the set of symbols used in text input to the model.
+'''
+_pad        = '_'
+_punctuation = ',.!?-~…'
+_letters = 'AEINOQUabdefghijklmnoprstuvwyzʃʧʦɯɹəɥ⁼ʰ`→↓↑ '
+'''
+# japanese_cleaners2
+_pad        = '_'
+_punctuation = ',.!?-~…'
+_letters = 'AEINOQUabdefghijkmnoprstuvwyzʃʧʦ↓↑ '
+'''
+
+'''# korean_cleaners
+_pad        = '_'
+_punctuation = ',.!?…~'
+_letters = 'ㄱㄴㄷㄹㅁㅂㅅㅇㅈㅊㅋㅌㅍㅎㄲㄸㅃㅆㅉㅏㅓㅗㅜㅡㅣㅐㅔ '
+'''
+
+'''# chinese_cleaners
+_pad        = '_'
+_punctuation = '，。！？—…'
+_letters = 'ㄅㄆㄇㄈㄉㄊㄋㄌㄍㄎㄏㄐㄑㄒㄓㄔㄕㄖㄗㄘㄙㄚㄛㄜㄝㄞㄟㄠㄡㄢㄣㄤㄥㄦㄧㄨㄩˉˊˇˋ˙ '
+'''
+
+
+'''# sanskrit_cleaners
+_pad        = '_'
+_punctuation = '।'
+_letters = 'ँंःअआइईउऊऋएऐओऔकखगघङचछजझञटठडढणतथदधनपफबभमयरलळवशषसहऽािीुूृॄेैोौ्ॠॢ '
+'''
+
+'''# cjks_cleaners
+_pad        = '_'
+_punctuation = ',.!?-~…'
+_letters = 'NQabdefghijklmnopstuvwxyzʃʧʥʦɯɹəɥçɸɾβŋɦː⁼ʰ`^#*=→↓↑ '
+'''
+
+'''# thai_cleaners
+_pad        = '_'
+_punctuation = '.!? '
+_letters = 'กขฃคฆงจฉชซฌญฎฏฐฑฒณดตถทธนบปผฝพฟภมยรฤลวศษสหฬอฮฯะัาำิีึืุูเแโใไๅๆ็่้๊๋์'
+'''
+
+'''# cjke_cleaners2
+_pad        = '_'
+_punctuation = ',.!?-~…'
+_letters = 'NQabdefghijklmnopstuvwxyzɑæʃʑçɯɪɔɛɹðəɫɥɸʊɾʒθβŋɦ⁼ʰ`^#*=ˈˌ→↓↑ '
+'''
+
+'''# shanghainese_cleaners
+_pad        = '_'
+_punctuation = ',.!?…'
+_letters = 'abdfghiklmnopstuvyzøŋȵɑɔɕəɤɦɪɿʑʔʰ̩̃ᴀᴇ15678 '
+'''
+
+'''# chinese_dialect_cleaners
+_pad        = '_'
+_punctuation = ',.!?~…─'
+_letters = '#Nabdefghijklmnoprstuvwxyzæçøŋœȵɐɑɒɓɔɕɗɘəɚɛɜɣɤɦɪɭɯɵɷɸɻɾɿʂʅʊʋʌʏʑʔʦʮʰʷˀː˥˦˧˨˩̥̩̃̚αᴀᴇ↑↓∅ⱼ '
+'''
+
+# Export all symbols:
+symbols = [_pad] + list(_punctuation) + list(_letters)
+
+# Special symbol ids
+SPACE_ID = symbols.index(" ")
diff --git a/text/thai.py b/text/thai.py
new file mode 100644
index 0000000000000000000000000000000000000000..998207c01a85c710a46db1ec8b62c39c2d94bc84
--- /dev/null
+++ b/text/thai.py
@@ -0,0 +1,44 @@
+import re
+from num_thai.thainumbers import NumThai
+
+
+num = NumThai()
+
+# List of (Latin alphabet, Thai) pairs:
+_latin_to_thai = [(re.compile('%s' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('a', 'เอ'),
+    ('b','บี'),
+    ('c','ซี'),
+    ('d','ดี'),
+    ('e','อี'),
+    ('f','เอฟ'),
+    ('g','จี'),
+    ('h','เอช'),
+    ('i','ไอ'),
+    ('j','เจ'),
+    ('k','เค'),
+    ('l','แอล'),
+    ('m','เอ็ม'),
+    ('n','เอ็น'),
+    ('o','โอ'),
+    ('p','พี'),
+    ('q','คิว'),
+    ('r','แอร์'),
+    ('s','เอส'),
+    ('t','ที'),
+    ('u','ยู'),
+    ('v','วี'),
+    ('w','ดับเบิลยู'),
+    ('x','เอ็กซ์'),
+    ('y','วาย'),
+    ('z','ซี')
+]]
+
+
+def num_to_thai(text):
+    return re.sub(r'(?:\d+(?:,?\d+)?)+(?:\.\d+(?:,?\d+)?)?', lambda x: ''.join(num.NumberToTextThai(float(x.group(0).replace(',', '')))), text)
+
+def latin_to_thai(text):
+    for regex, replacement in _latin_to_thai:
+        text = re.sub(regex, replacement, text)
+    return text
diff --git a/transforms.py b/transforms.py
new file mode 100644
index 0000000000000000000000000000000000000000..30e857692442329a569fdd53e391b3ec4044a628
--- /dev/null
+++ b/transforms.py
@@ -0,0 +1,200 @@
+import numpy as np
+import torch
+from torch.nn import functional as F
+
+DEFAULT_MIN_BIN_WIDTH = 1e-3
+DEFAULT_MIN_BIN_HEIGHT = 1e-3
+DEFAULT_MIN_DERIVATIVE = 1e-3
+
+
+def piecewise_rational_quadratic_transform(
+        inputs,
+        unnormalized_widths,
+        unnormalized_heights,
+        unnormalized_derivatives,
+        inverse=False,
+        tails=None,
+        tail_bound=1.,
+        min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+        min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+        min_derivative=DEFAULT_MIN_DERIVATIVE):
+
+    if tails is None:
+        spline_fn = rational_quadratic_spline
+        spline_kwargs = {}
+    else:
+        spline_fn = unconstrained_rational_quadratic_spline
+        spline_kwargs = {'tails': tails, 'tail_bound': tail_bound}
+
+    outputs, logabsdet = spline_fn(
+        inputs=inputs,
+        unnormalized_widths=unnormalized_widths,
+        unnormalized_heights=unnormalized_heights,
+        unnormalized_derivatives=unnormalized_derivatives,
+        inverse=inverse,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+        **spline_kwargs)
+    return outputs, logabsdet
+
+
+def searchsorted(bin_locations, inputs, eps=1e-6):
+    bin_locations[..., bin_locations.size(-1) - 1] += eps
+    return torch.sum(inputs[..., None] >= bin_locations, dim=-1) - 1
+
+
+def unconstrained_rational_quadratic_spline(
+        inputs,
+        unnormalized_widths,
+        unnormalized_heights,
+        unnormalized_derivatives,
+        inverse=False,
+        tails='linear',
+        tail_bound=1.,
+        min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+        min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+        min_derivative=DEFAULT_MIN_DERIVATIVE):
+    inside_interval_mask = (inputs >= -tail_bound) & (inputs <= tail_bound)
+    outside_interval_mask = ~inside_interval_mask
+
+    outputs = torch.zeros_like(inputs)
+    logabsdet = torch.zeros_like(inputs)
+
+    if tails == 'linear':
+        unnormalized_derivatives = F.pad(unnormalized_derivatives, pad=(1, 1))
+        constant = np.log(np.exp(1 - min_derivative) - 1)
+        unnormalized_derivatives[..., 0] = constant
+        unnormalized_derivatives[..., unnormalized_derivatives.size(-1) - 1] = constant
+
+        outputs[outside_interval_mask] = inputs[outside_interval_mask]
+        logabsdet[outside_interval_mask] = 0
+    else:
+        raise RuntimeError('{} tails are not implemented.'.format(tails))
+
+    outputs[inside_interval_mask], logabsdet[
+        inside_interval_mask] = rational_quadratic_spline(
+            inputs=inputs[inside_interval_mask],
+            unnormalized_widths=unnormalized_widths[inside_interval_mask, :],
+            unnormalized_heights=unnormalized_heights[inside_interval_mask, :],
+            unnormalized_derivatives=unnormalized_derivatives[
+                inside_interval_mask, :],
+            inverse=inverse,
+            left=-tail_bound,
+            right=tail_bound,
+            bottom=-tail_bound,
+            top=tail_bound,
+            min_bin_width=min_bin_width,
+            min_bin_height=min_bin_height,
+            min_derivative=min_derivative)
+
+    return outputs, logabsdet
+
+
+def rational_quadratic_spline(inputs,
+                              unnormalized_widths,
+                              unnormalized_heights,
+                              unnormalized_derivatives,
+                              inverse=False,
+                              left=0.,
+                              right=1.,
+                              bottom=0.,
+                              top=1.,
+                              min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+                              min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+                              min_derivative=DEFAULT_MIN_DERIVATIVE):
+    if torch.min(inputs) < left or torch.max(inputs) > right:
+        raise ValueError('Input to a transform is not within its domain')
+
+    num_bins = unnormalized_widths.shape[-1]
+
+    if min_bin_width * num_bins > 1.0:
+        raise ValueError('Minimal bin width too large for the number of bins')
+    if min_bin_height * num_bins > 1.0:
+        raise ValueError('Minimal bin height too large for the number of bins')
+
+    widths = F.softmax(unnormalized_widths, dim=-1)
+    widths = min_bin_width + (1 - min_bin_width * num_bins) * widths
+    cumwidths = torch.cumsum(widths, dim=-1)
+    cumwidths = F.pad(cumwidths, pad=(1, 0), mode='constant', value=0.0)
+    cumwidths = (right - left) * cumwidths + left
+    cumwidths[..., 0] = left
+    cumwidths[..., cumwidths.size(-1) - 1] = right
+    widths = cumwidths[..., 1:] - cumwidths[..., :-1]
+
+    derivatives = min_derivative + F.softplus(unnormalized_derivatives)
+
+    heights = F.softmax(unnormalized_heights, dim=-1)
+    heights = min_bin_height + (1 - min_bin_height * num_bins) * heights
+    cumheights = torch.cumsum(heights, dim=-1)
+    cumheights = F.pad(cumheights, pad=(1, 0), mode='constant', value=0.0)
+    cumheights = (top - bottom) * cumheights + bottom
+    cumheights[..., 0] = bottom
+    cumheights[..., cumheights.size(-1) - 1] = top
+    heights = cumheights[..., 1:] - cumheights[..., :-1]
+
+    if inverse:
+        bin_idx = searchsorted(cumheights, inputs)[..., None]
+    else:
+        bin_idx = searchsorted(cumwidths, inputs)[..., None]
+
+    input_cumwidths = cumwidths.gather(-1, bin_idx)[..., 0]
+    input_bin_widths = widths.gather(-1, bin_idx)[..., 0]
+
+    input_cumheights = cumheights.gather(-1, bin_idx)[..., 0]
+    delta = heights / widths
+    input_delta = delta.gather(-1, bin_idx)[..., 0]
+
+    input_derivatives = derivatives.gather(-1, bin_idx)[..., 0]
+    input_derivatives_plus_one = derivatives[..., 1:].gather(-1, bin_idx)[...,
+                                                                          0]
+
+    input_heights = heights.gather(-1, bin_idx)[..., 0]
+
+    if inverse:
+        a = (
+            ((inputs - input_cumheights) *
+             (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+             + input_heights * (input_delta - input_derivatives)))
+        b = (
+            input_heights * input_derivatives - (inputs - input_cumheights) *
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta))
+        c = -input_delta * (inputs - input_cumheights)
+
+        discriminant = b.pow(2) - 4 * a * c
+        assert (discriminant >= 0).all()
+
+        root = (2 * c) / (-b - torch.sqrt(discriminant))
+        outputs = root * input_bin_widths + input_cumwidths
+
+        theta_one_minus_theta = root * (1 - root)
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta)
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * root.pow(2) +
+            2 * input_delta * theta_one_minus_theta + input_derivatives *
+            (1 - root).pow(2))
+        logabsdet = torch.log(
+            derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, -logabsdet
+    else:
+        theta = (inputs - input_cumwidths) / input_bin_widths
+        theta_one_minus_theta = theta * (1 - theta)
+
+        numerator = input_heights * (input_delta * theta.pow(2) +
+                                     input_derivatives * theta_one_minus_theta)
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta)
+        outputs = input_cumheights + numerator / denominator
+
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * theta.pow(2) +
+            2 * input_delta * theta_one_minus_theta + input_derivatives *
+            (1 - theta).pow(2))
+        logabsdet = torch.log(
+            derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, logabsdet
diff --git a/utils.py b/utils.py
new file mode 100644
index 0000000000000000000000000000000000000000..a870556e805c8cba7dc0540c686710a6a62819c4
--- /dev/null
+++ b/utils.py
@@ -0,0 +1,307 @@
+import argparse
+import glob
+import json
+import logging
+import os
+import subprocess
+import sys
+
+import numpy as np
+from scipy.io.wavfile import read
+import torch
+
+MATPLOTLIB_FLAG = False
+
+logging.basicConfig(stream=sys.stdout, level=logging.INFO)
+logger = logging
+
+
+def load_checkpoint(checkpoint_path, model, optimizer=None):
+    assert os.path.isfile(checkpoint_path)
+    checkpoint_dict = torch.load(checkpoint_path, map_location='cpu')
+    iteration = checkpoint_dict['iteration']
+    learning_rate = checkpoint_dict['learning_rate']
+    if optimizer is not None:
+        optimizer.load_state_dict(checkpoint_dict['optimizer'])
+    saved_state_dict = checkpoint_dict['model']
+    if hasattr(model, 'module'):
+        state_dict = model.module.state_dict()
+    else:
+        state_dict = model.state_dict()
+    new_state_dict = {}
+    for k, v in state_dict.items():
+        try:
+            new_state_dict[k] = saved_state_dict[k]
+        except Exception as e:
+            logger.info("%s is not in the checkpoint" % k)
+            new_state_dict[k] = v
+    if hasattr(model, 'module'):
+        model.module.load_state_dict(new_state_dict)
+    else:
+        model.load_state_dict(new_state_dict)
+    logger.info("Loaded checkpoint '{}' (iteration {})".format(
+        checkpoint_path, iteration))
+    return model, optimizer, learning_rate, iteration
+
+
+def save_checkpoint(model, optimizer, learning_rate, iteration,
+                    checkpoint_path):
+    logger.info(
+        "Saving model and optimizer state at iteration {} to {}".format(
+            iteration, checkpoint_path))
+    if hasattr(model, 'module'):
+        state_dict = model.module.state_dict()
+    else:
+        state_dict = model.state_dict()
+    torch.save(
+        {
+            'model': state_dict,
+            'iteration': iteration,
+            'optimizer': optimizer.state_dict(),
+            'learning_rate': learning_rate
+        }, checkpoint_path)
+
+
+def summarize(
+        writer,
+        global_step,
+        scalars={},  # noqa
+        histograms={},  # noqa
+        images={},  # noqa
+        audios={},  # noqa
+        audio_sampling_rate=22050):
+    for k, v in scalars.items():
+        writer.add_scalar(k, v, global_step)
+    for k, v in histograms.items():
+        writer.add_histogram(k, v, global_step)
+    for k, v in images.items():
+        writer.add_image(k, v, global_step, dataformats='HWC')
+    for k, v in audios.items():
+        writer.add_audio(k, v, global_step, audio_sampling_rate)
+
+
+def latest_checkpoint_path(dir_path, regex="G_*.pth"):
+    f_list = glob.glob(os.path.join(dir_path, regex))
+    f_list.sort(key=lambda f: int("".join(filter(str.isdigit, f))))
+    x = f_list[-1]
+    print(x)
+    return x
+
+
+def plot_spectrogram_to_numpy(spectrogram):
+    global MATPLOTLIB_FLAG
+    if not MATPLOTLIB_FLAG:
+        import matplotlib
+        matplotlib.use("Agg")
+        MATPLOTLIB_FLAG = True
+        mpl_logger = logging.getLogger('matplotlib')
+        mpl_logger.setLevel(logging.WARNING)
+    import matplotlib.pylab as plt
+    import numpy as np
+
+    fig, ax = plt.subplots(figsize=(10, 2))
+    im = ax.imshow(spectrogram,
+                   aspect="auto",
+                   origin="lower",
+                   interpolation='none')
+    plt.colorbar(im, ax=ax)
+    plt.xlabel("Frames")
+    plt.ylabel("Channels")
+    plt.tight_layout()
+
+    fig.canvas.draw()
+    data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')
+    data = data.reshape(fig.canvas.get_width_height()[::-1] + (3, ))
+    plt.close()
+    return data
+
+
+def plot_alignment_to_numpy(alignment, info=None):
+    global MATPLOTLIB_FLAG
+    if not MATPLOTLIB_FLAG:
+        import matplotlib
+        matplotlib.use("Agg")
+        MATPLOTLIB_FLAG = True
+        mpl_logger = logging.getLogger('matplotlib')
+        mpl_logger.setLevel(logging.WARNING)
+    import matplotlib.pylab as plt
+    import numpy as np
+
+    fig, ax = plt.subplots(figsize=(6, 4))
+    im = ax.imshow(alignment.transpose(),
+                   aspect='auto',
+                   origin='lower',
+                   interpolation='none')
+    fig.colorbar(im, ax=ax)
+    xlabel = 'Decoder timestep'
+    if info is not None:
+        xlabel += '\n\n' + info
+    plt.xlabel(xlabel)
+    plt.ylabel('Encoder timestep')
+    plt.tight_layout()
+
+    fig.canvas.draw()
+    data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')
+    data = data.reshape(fig.canvas.get_width_height()[::-1] + (3, ))
+    plt.close()
+    return data
+
+
+def load_wav_to_torch(full_path):
+    sampling_rate, data = read(full_path)
+    return torch.FloatTensor(data.astype(np.float32)), sampling_rate
+
+
+def load_filepaths_and_text(filename, split="|"):
+    with open(filename, encoding='utf-8') as f:
+        filepaths_and_text = [line.strip().split(split) for line in f]
+    return filepaths_and_text
+
+
+def get_hparams(init=True):
+    parser = argparse.ArgumentParser()
+    parser.add_argument('-c',
+                        '--config',
+                        type=str,
+                        default="./configs/base.json",
+                        help='JSON file for configuration')
+    parser.add_argument('-m',
+                        '--model',
+                        type=str,
+                        required=True,
+                        help='Model name')
+    parser.add_argument('--train_data',
+                        type=str,
+                        required=True,
+                        help='train data')
+    parser.add_argument('--val_data', type=str, required=True, help='val data')
+    parser.add_argument('--phone_table',
+                        type=str,
+                        required=True,
+                        help='phone table')
+    parser.add_argument('--speaker_table',
+                        type=str,
+                        default=None,
+                        help='speaker table, required for multiple speakers')
+
+    args = parser.parse_args()
+    model_dir = args.model
+
+    if not os.path.exists(model_dir):
+        os.makedirs(model_dir)
+
+    config_path = args.config
+    config_save_path = os.path.join(model_dir, "config.json")
+    if init:
+        with open(config_path, "r", encoding='utf8') as f:
+            data = f.read()
+        with open(config_save_path, "w", encoding='utf8') as f:
+            f.write(data)
+    else:
+        with open(config_save_path, "r", encoding='utf8') as f:
+            data = f.read()
+    config = json.loads(data)
+    config['data']['training_files'] = args.train_data
+    config['data']['validation_files'] = args.val_data
+    config['data']['phone_table'] = args.phone_table
+    # 0 is kept for blank
+    config['data']['num_phones'] = len(open(args.phone_table).readlines()) + 1
+    if args.speaker_table is not None:
+        config['data']['speaker_table'] = args.speaker_table
+        # 0 is kept for unknown speaker
+        config['data']['n_speakers'] = len(
+            open(args.speaker_table).readlines()) + 1
+    else:
+        config['data']['n_speakers'] = 0
+
+    hparams = HParams(**config)
+    hparams.model_dir = model_dir
+    return hparams
+
+
+def get_hparams_from_dir(model_dir):
+    config_save_path = os.path.join(model_dir, "config.json")
+    with open(config_save_path, "r") as f:
+        data = f.read()
+    config = json.loads(data)
+
+    hparams = HParams(**config)
+    hparams.model_dir = model_dir
+    return hparams
+
+
+def get_hparams_from_file(config_path):
+    with open(config_path, "r") as f:
+        data = f.read()
+    config = json.loads(data)
+
+    hparams = HParams(**config)
+    return hparams
+
+
+def check_git_hash(model_dir):
+    source_dir = os.path.dirname(os.path.realpath(__file__))
+    if not os.path.exists(os.path.join(source_dir, ".git")):
+        logger.warn('''{} is not a git repository, therefore hash value
+                       comparison will be ignored.'''.format(source_dir))
+        return
+
+    cur_hash = subprocess.getoutput("git rev-parse HEAD")
+
+    path = os.path.join(model_dir, "githash")
+    if os.path.exists(path):
+        saved_hash = open(path).read()
+        if saved_hash != cur_hash:
+            logger.warn(
+                "git hash values are different. {}(saved) != {}(current)".
+                format(saved_hash[:8], cur_hash[:8]))
+    else:
+        open(path, "w").write(cur_hash)
+
+
+def get_logger(model_dir, filename="train.log"):
+    global logger
+    logger = logging.getLogger(os.path.basename(model_dir))
+    logger.setLevel(logging.INFO)
+
+    formatter = logging.Formatter(
+        "%(asctime)s\t%(name)s\t%(levelname)s\t%(message)s")
+    if not os.path.exists(model_dir):
+        os.makedirs(model_dir)
+    h = logging.FileHandler(os.path.join(model_dir, filename))
+    h.setLevel(logging.INFO)
+    h.setFormatter(formatter)
+    logger.addHandler(h)
+    return logger
+
+
+class HParams():
+    def __init__(self, **kwargs):
+        for k, v in kwargs.items():
+            if type(v) == dict:
+                v = HParams(**v)
+            self[k] = v
+
+    def keys(self):
+        return self.__dict__.keys()
+
+    def items(self):
+        return self.__dict__.items()
+
+    def values(self):
+        return self.__dict__.values()
+
+    def __len__(self):
+        return len(self.__dict__)
+
+    def __getitem__(self, key):
+        return getattr(self, key)
+
+    def __setitem__(self, key, value):
+        return setattr(self, key, value)
+
+    def __contains__(self, key):
+        return key in self.__dict__
+
+    def __repr__(self):
+        return self.__dict__.__repr__()