tokenizer_test.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !pip install tokenizers\n",
    "# !pip install -U sentencepiece"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 211,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "from pathlib import Path\n",
    "import jieba\n",
    "import sentencepiece as spm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 202,
   "metadata": {},
   "outputs": [],
   "source": [
    "jieba.initialize()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 203,
   "metadata": {},
   "outputs": [],
   "source": [
    "root = Path('CPM-Generate/bpe_3w_new/')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 213,
   "metadata": {},
   "outputs": [],
   "source": [
    "word_index = json.load(open(root / 'vocab.json'))\n",
    "index_word = {v: k for k, v in word_index.items()}\n",
    "spp = spm.SentencePieceProcessor(model_file=str(root / 'chinese_vocab.model'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 210,
   "metadata": {},
   "outputs": [],
   "source": [
    "def encode(s):\n",
    "    s = [x.translate(str.maketrans(\" \\n\", \"\\u2582\\u2583\")) for x in jieba.cut(s)]\n",
    "    s = ' '.join(s)\n",
    "    return spp.encode(s)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 219,
   "metadata": {},
   "outputs": [],
   "source": [
    "def decode(t):\n",
    "    text = [\n",
    "        index_word[x].replace(' ', '').replace('\\u2582', ' ').replace('\\u2583', '\\n').replace('\\u2584', '<eod>')\n",
    "        for x in t\n",
    "    ]\n",
    "    text = [\n",
    "        spp.decode(x)\n",
    "        for x in text\n",
    "    ]\n",
    "    return text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 220,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[25159, 16]"
      ]
     },
     "execution_count": 220,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "encode('hello我')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 221,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['hello', '我']"
      ]
     },
     "execution_count": 221,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "decode(encode('hello我'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}