forked from cocktailpeanut/dalai
-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.js
211 lines (206 loc) · 5.86 KB
/
index.js
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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
const os = require('os');
const pty = require('node-pty');
const path = require('path');
const fs = require("fs");
const { createServer } = require("http");
const { Server } = require("socket.io");
const { io } = require("socket.io-client");
const term = require( 'terminal-kit' ).terminal;
const Downloader = require("nodejs-file-downloader");
const shell = os.platform() === 'win32' ? 'powershell.exe' : 'bash';
class Dalai {
constructor(url) {
if (url) this.url = url
this.home = path.resolve(os.homedir(), "dalai")
try {
fs.mkdirSync(this.home, { recursive: true })
} catch (e) { }
this.config = {
name: 'xterm-color',
cols: 80,
rows: 30,
}
}
async download(model) {
const num = {
"7B": 1,
"13B": 2,
"30B": 4,
"65B": 8,
}
const files = ["checklist.chk", "params.json"]
for(let i=0; i<num[model]; i++) {
files.push(`consolidated.0${i}.pth`)
}
const resolvedPath = path.resolve(this.home, "models", model)
await fs.promises.mkdir(resolvedPath, { recursive: true }).catch((e) => { })
for(let file of files) {
const task = `downloading ${file}`
const downloader = new Downloader({
url: `https://agi.gpt4.org/llama/LLaMA/${model}/${file}`,
directory: path.resolve(this.home, "models", model),
onProgress: (percentage, chunk, remainingSize) => {
this.progress(task, percentage)
},
});
try {
await this.startProgress(task)
await downloader.download();
} catch (error) {
console.log(error);
}
this.progressBar.update(1);
term("\n")
}
const files2 = ["tokenizer_checklist.chk", "tokenizer.model"]
for(let file of files2) {
const task = `downloading ${file}`
const downloader = new Downloader({
url: `https://agi.gpt4.org/llama/LLaMA/${file}`,
directory: path.resolve(this.home, "models"),
onProgress: (percentage, chunk, remainingSize) => {
this.progress(task, percentage)
},
});
try {
await this.startProgress(task)
await downloader.download();
} catch (error) {
console.log(error);
}
this.progressBar.update(1);
term("\n")
}
}
async install(...models) {
// install to ~/llama.cpp
await this.exec("pip3 install torch torchvision torchaudio sentencepiece numpy")
await this.exec("pip install torch torchvision torchaudio sentencepiece numpy")
await this.exec("git clone https://github.com/ggerganov/llama.cpp.git dalai", os.homedir())
await this.exec("make", this.home)
for(let model of models) {
await this.download(model)
await this.exec(`python3 convert-pth-to-ggml.py models/${model}/ 1`, this.home)
await this.quantize(model)
}
}
serve(port) {
const httpServer = createServer();
const io = new Server(httpServer)
io.on("connection", (socket) => {
socket.on('request', async (req) => {
await this.query(req, (str) => {
io.emit("result", { response: str, request: req })
})
});
});
httpServer.listen(port)
}
http(httpServer) {
const io = new Server(httpServer)
io.on("connection", (socket) => {
socket.on('request', async (req) => {
await this.query(req, (str) => {
io.emit("result", { response: str, request: req })
})
});
});
}
async request(req, cb) {
if (this.url) {
await this.connect(req, cb)
} else {
await this.query(req, cb)
}
}
async query(req, cb) {
let o = {
seed: req.seed || -1,
threads: req.threads || 8,
n_predict: req.n_predict || 128,
model: `./models/${req.model || "7B"}/ggml-model-q4_0.bin`
}
if (req.top_k) o.top_k = req.top_k
if (req.top_p) o.top_p = req.top_p
if (req.temp) o.temp = req.temp
if (req.batch_size) o.batch_size = req.batch_size
let chunks = []
for(let key in o) {
chunks.push(`--${key} ${o[key]}`)
}
chunks.push(`-p "${req.prompt}"`)
if (req.full) {
await this.exec(`./main ${chunks.join(" ")}`, this.home, cb)
} else {
const startpattern = /.*sampling parameters:.*/g
const endpattern = /.*mem per token.*/g
let started = false
let ended = false
await this.exec(`./main ${chunks.join(" ")}`, this.home, (msg) => {
if (endpattern.test(msg)) ended = true
if (started && !ended) {
cb(msg)
}
if (startpattern.test(msg)) started = true
})
}
}
connect(req, cb) {
const socket = io(this.url)
socket.emit('request', req)
socket.on('response', cb)
socket.on('error', function(e) {
throw e
});
}
exec(cmd, cwd, cb) {
return new Promise((resolve, reject) => {
const config = Object.assign({}, this.config)
if (cwd) {
config.cwd = path.resolve(cwd)
}
const ptyProcess = pty.spawn(shell, [], config)
ptyProcess.onData((data) => {
if (cb) {
cb(data)
} else {
process.stdout.write(data);
}
});
ptyProcess.onExit((res) => {
resolve(res)
});
ptyProcess.write(`${cmd}\r`)
ptyProcess.write("exit\r")
})
}
async quantize(model) {
let num = {
"7B": 1,
"13B": 2,
"30B": 4,
"65B": 8,
}
for(let i=0; i<num[model]; i++) {
const suffix = (i === 0 ? "" : `.${i}`)
await this.exec(`./quantize ./models/${model}/ggml-model-f16.bin ./models/${model}/ggml-model-q4_0.bin${suffix} 2`, this.home)
}
}
progress(task, percent) {
this.progressBar.update(percent/100);
//if (percent >= 100) {
// setTimeout(() => {
// term("\n")
// }, 200)
//}
}
startProgress(title) {
this.progressBar = term.progressBar({
width: 120,
title,
eta: true ,
percent: true
});
}
}
module.exports = Dalai