File size: 3,972 Bytes
2f35054
415aaef
2f35054
 
 
 
 
 
 
ac2af95
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2f35054
 
 
 
 
 
 
 
 
 
 
 
 
ca67cfa
 
2f35054
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ca67cfa
 
 
 
 
2f35054
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
/* eslint-disable no-restricted-globals */
import { pipeline } from 'https://cdn.jsdelivr.net/npm/@huggingface/transformers@latest'

class MyTextGenerationPipeline {
  static task = 'text-generation'
  static instance = null
  static currentGeneration = null

  static async getInstance(model, dtype = 'fp32', progress_callback = null) {
    try {
      // Try WebGPU first
      this.instance = await pipeline(this.task, model, {
        dtype,
        device: 'webgpu',
        progress_callback
      })
      return this.instance
    } catch (webgpuError) {
      // Fallback to WASM if WebGPU fails
      if (progress_callback) {
        progress_callback({
          status: 'fallback',
          message: 'WebGPU failed, falling back to WASM'
        })
      }
      try {
        this.instance = await pipeline(this.task, model, {
          dtype,
          device: 'wasm',
          progress_callback
        })
        return this.instance
      } catch (wasmError) {
        throw new Error(
          `Both WebGPU and WASM failed. WebGPU error: ${webgpuError.message}. WASM error: ${wasmError.message}`
        )
      }
    }
  }

  static stopGeneration() {
    if (this.currentGeneration) {
      this.currentGeneration.abort()
      this.currentGeneration = null
    }
  }
}

// Listen for messages from the main thread
self.addEventListener('message', async (event) => {
  try {
    const { type, model, dtype, messages, prompt, hasChatTemplate, config } =
      event.data

    if (type === 'stop') {
      MyTextGenerationPipeline.stopGeneration()
      self.postMessage({ status: 'ready' })
      return
    }

    if (!model) {
      self.postMessage({
        status: 'error',
        output: 'No model provided'
      })
      return
    }

    // Retrieve the pipeline. This will download the model if not already cached.
    const generator = await MyTextGenerationPipeline.getInstance(
      model,
      dtype,
      (x) => {
        self.postMessage({ status: 'loading', output: x })
      }
    )

    if (type === 'load') {
      self.postMessage({
        status: 'ready',
        output: `Model ${model}, dtype ${dtype} loaded`
      })
      return
    }

    if (type === 'generate') {
      let inputText = ''

      if (hasChatTemplate && messages && messages.length > 0) {
        inputText = messages
      } else if (!hasChatTemplate && prompt) {
        inputText = prompt
      } else {
        self.postMessage({ status: 'ready' })
        return
      }

      const options = {
        max_new_tokens: config.max_new_tokens || 100,
        temperature: config.temperature || 0.7,
        do_sample: config.do_sample !== false,
        ...(config.top_p && { top_p }),
        ...(config.top_k && { top_k })
      }

      // Create an AbortController for this generation
      const abortController = new AbortController()
      MyTextGenerationPipeline.currentGeneration = abortController

      try {
        const output = await generator(inputText, {
          ...options,
          signal: abortController.signal
        })

        if (hasChatTemplate) {
          // For chat mode, extract only the assistant's response
          self.postMessage({
            status: 'output',
            output: output[0].generated_text.slice(-1)[0]
          })
        } else {
          self.postMessage({
            status: 'output',
            output: {
              role: 'assistant',
              content: output[0].generated_text
            }
          })
        }

        self.postMessage({ status: 'ready' })
      } catch (error) {
        if (error.name === 'AbortError') {
          self.postMessage({ status: 'ready' })
        } else {
          throw error
        }
      } finally {
        MyTextGenerationPipeline.currentGeneration = null
      }
    }
  } catch (error) {
    self.postMessage({
      status: 'error',
      output: error.message || 'An error occurred during text generation'
    })
  }
})