File size: 6,265 Bytes
b092c58
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "54afb2a8",
   "metadata": {},
   "source": [
    "# Generate a single [character.ai](https://beta.character.ai/) character definition\n",
    "\n",
    "This example shows how to generate the character definition of a single [character.ai](https://beta.character.ai/) character from a corpus. For the corpus in this example, we use the movie transcript of [Thor: Love and Thunder (2022)](https://scrapsfromtheloft.com/movies/thor-love-and-thunder-transcript/).\n",
    "\n",
    "To generate your own character definition:\n",
    "1. Put the corpus into a single a `.txt` file inside the `data/` directory.\n",
    "2. Assign the name of the `.txt` file to the `CORPUS` constant below.\n",
    "3. Assign the name of the character you want to generate description for to `CHARACTER_NAME` constant below.\n",
    "4. Run this notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "2c5d195f",
   "metadata": {},
   "outputs": [],
   "source": [
    "CORPUS = 'data/thor_love_and_thunder.txt'\n",
    "CHARACTER_NAME = \"Jane Foster\"  # the name of the character we want to generate a description for"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "da765a49",
   "metadata": {},
   "outputs": [],
   "source": [
    "from dataclasses import asdict\n",
    "import json\n",
    "import os\n",
    "\n",
    "from data_driven_characters.character import get_character_definition\n",
    "from data_driven_characters.corpus import get_characters, get_corpus_summaries, load_docs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "8298d68b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# create directories to cache results and intermediate outputs\n",
    "SUMMARY_TYPE = \"map_reduce\"  # summarize each chunk of the corpus independently\n",
    "OUTPUT_ROOT = \"output\"\n",
    "corpus_name = os.path.splitext(os.path.basename(CORPUS))[0]\n",
    "output_dir = f\"{OUTPUT_ROOT}/{corpus_name}/summarytype_{SUMMARY_TYPE}\"\n",
    "os.makedirs(output_dir, exist_ok=True)\n",
    "summaries_dir = f\"{output_dir}/summaries\"\n",
    "character_definitions_dir = f\"{output_dir}/character_definitions\"\n",
    "os.makedirs(character_definitions_dir, exist_ok=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "baf9e861",
   "metadata": {},
   "source": [
    "## Summarization\n",
    "Because the entire corpus does not fit in the context length of the LLM, we split it into a list of chunks.\n",
    "We turn the list of chunks into a list of summaries using one of [LangChain's summarization chains](https://langchain-langchain.vercel.app/docs/modules/chains/document/).\n",
    "\n",
    "If `SUMMARY_TYPE = 'refine'`, we first summarize the first chunk, and then each subsequent summary is generated from the previous summary and the current chunk.\n",
    "If `SUMMARY_TYPE = 'map_reduce'`, we summarize each chunk independently.\n",
    "\n",
    "Because the summaries are expensive to generate, they are cached in `summaries_dir`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "f72b8d1c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Summaries already exist. Loading summaries.\n"
     ]
    }
   ],
   "source": [
    "# split corpus into a set of chunks\n",
    "docs = load_docs(\n",
    "    corpus_path=CORPUS,\n",
    "    chunk_size=2048,  # number of tokens per chunk\n",
    "    chunk_overlap=64,  # number of tokens of overlap between chunks\n",
    ")\n",
    "\n",
    "# generate summaries\n",
    "corpus_summaries = get_corpus_summaries(docs=docs, summary_type=SUMMARY_TYPE, cache_dir=summaries_dir)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a0f116f3",
   "metadata": {},
   "source": [
    "## Generate [character.ai](https://beta.character.ai/) character definition\n",
    "Based on the corpus, we can now generate the elements - name, short description (50 characters), long description (500 characters), and custom greeting - that are required to [create a character.ai character](https://beta.character.ai/editing). These character definitions are cached in `character_definitions_dir`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "45d827ce",
   "metadata": {},
   "outputs": [],
   "source": [
    "character_definition = get_character_definition(\n",
    "        name=CHARACTER_NAME,\n",
    "        corpus_summaries=corpus_summaries,\n",
    "        cache_dir=character_definitions_dir,\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "ce604024",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "    \"name\": \"Jane Foster\",\n",
      "    \"short_description\": \"I'm Jane Foster, fighting cancer & evil.\",\n",
      "    \"long_description\": \"I am Jane Foster, a determined woman battling stage four cancer, yet fighting alongside Thor against the evil God Butcher, Gorr. My deep connection with Thor strengthens our resolve. As the Mighty Thor, I wield Mjolnir, despite its draining effect. Fiercely independent, I refuse help from close friends. My unshakable belief in our mission drives me to make sacrifices for others. Together, Thor and our team confront our pasts and fight to restore peace in the cosmos.\",\n",
      "    \"greeting\": \"Hi there, I'm Jane. Ready to take on whatever challenges come our way?\"\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "print(json.dumps(asdict(character_definition), indent=4))"
   ]
  }
 ],
 "metadata": {
  "jupytext": {
   "cell_metadata_filter": "-all",
   "main_language": "python",
   "notebook_metadata_filter": "-all"
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}