File size: 4,050 Bytes
31b13da
 
 
 
 
 
 
 
 
 
 
 
 
 
f9fe51a
 
 
 
 
 
 
 
 
 
 
 
 
 
31b13da
 
 
051c520
 
 
 
 
 
 
31b13da
 
 
 
 
051c520
 
f9fe51a
051c520
 
 
31b13da
 
 
 
051c520
31b13da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
051c520
31b13da
 
 
 
 
 
 
 
 
 
 
 
 
 
051c520
31b13da
 
 
 
b48dd87
31b13da
 
 
 
 
 
 
 
 
b48dd87
31b13da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import init, { Model } from "./build/m.js";

async function fetchArrayBuffer(url) {
  const cacheName = "phi-mixformer-candle-cache";
  const cache = await caches.open(cacheName);
  const cachedResponse = await cache.match(url);
  if (cachedResponse) {
    const data = await cachedResponse.arrayBuffer();
    return new Uint8Array(data);
  }
  const res = await fetch(url, { cache: "force-cache" });
  cache.put(url, res.clone());
  return new Uint8Array(await res.arrayBuffer());
}
async function concatenateArrayBuffers(urls) {
  const arrayBuffers = await Promise.all(urls.map(url => fetchArrayBuffer(url)));

  let totalLength = arrayBuffers.reduce((acc, arrayBuffer) => acc + arrayBuffer.byteLength, 0);
  let concatenatedBuffer = new Uint8Array(totalLength);

  let offset = 0;
  arrayBuffers.forEach(buffer => {
    concatenatedBuffer.set(new Uint8Array(buffer), offset);
    offset += buffer.byteLength;
  });
  return concatenatedBuffer;
}

class Phi {
  static instance = {};

  static async getInstance(
    weightsURL,
    modelID,
    tokenizerURL,
    configURL,
    quantized
  ) {
    // load individual modelID only once
    if (!this.instance[modelID]) {
      await init();

      self.postMessage({ status: "loading", message: "Loading Model" });
      const [weightsArrayU8, tokenizerArrayU8, configArrayU8] =
        await Promise.all([
          weightsURL instanceof Array ? concatenateArrayBuffers(weightsURL) : fetchArrayBuffer(weightsURL),
          fetchArrayBuffer(tokenizerURL),
          fetchArrayBuffer(configURL),
        ]);

      this.instance[modelID] = new Model(
        weightsArrayU8,
        tokenizerArrayU8,
        configArrayU8,
        quantized
      );
    }
    return this.instance[modelID];
  }
}

let controller = null;
self.addEventListener("message", (event) => {
  if (event.data.command === "start") {
    controller = new AbortController();
    generate(event.data);
  } else if (event.data.command === "abort") {
    controller.abort();
  }
});

async function generate(data) {
  const {
    weightsURL,
    modelID,
    tokenizerURL,
    configURL,
    quantized,
    prompt,
    temp,
    top_p,
    repeatPenalty,
    seed,
    maxSeqLen,
  } = data;
  try {
    self.postMessage({ status: "loading", message: "Starting Phi" });
    const model = await Phi.getInstance(
      weightsURL,
      modelID,
      tokenizerURL,
      configURL,
      quantized
    );

    self.postMessage({ status: "loading", message: "Initializing model" });
    const firstToken = model.init_with_prompt(
      prompt,
      temp,
      top_p,
      repeatPenalty,
      64,
      BigInt(seed)
    );
    const seq_len = 2048;

    let sentence = firstToken;
    let maxTokens = maxSeqLen ? maxSeqLen : seq_len - prompt.length - 1;
    let startTime = performance.now();
    let tokensCount = 0;
    while (tokensCount < maxTokens) {
      await new Promise(async (resolve) => {
        if (controller && controller.signal.aborted) {
          self.postMessage({
            status: "aborted",
            message: "Aborted",
            output: prompt + sentence,
          });
          return;
        }
        const token = await model.next_token();
        if (token === "<|endoftext|>") {
          self.postMessage({
            status: "complete",
            message: "complete",
            output: prompt + sentence,
          });
          return;
        }
        const tokensSec =
          ((tokensCount + 1) / (performance.now() - startTime)) * 1000;

        sentence += token;
        self.postMessage({
          status: "generating",
          message: "Generating token",
          token: token,
          sentence: sentence,
          totalTime: performance.now() - startTime,
          tokensSec,
          prompt: prompt,
        });
        setTimeout(resolve, 0);
      });
      tokensCount++;
    }
    self.postMessage({
      status: "complete",
      message: "complete",
      output: prompt + sentence,
    });
  } catch (e) {
    self.postMessage({ error: e });
  }
}