onnx_javascript_browser_inference/web.html at main · code2k13/onnx_javascript_browser_inference · GitHub

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
<!DOCTYPE html>
<html lang="en">

<head>
    <meta charset="UTF-8">
    <title>🪄 ONNX Paraphraser (Local)</title>
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <style>
        body {
            font-family: system-ui, sans-serif;
            background: #fafafa;
            max-width: 600px;
            margin: 40px auto;
            padding: 20px;
            border-radius: 12px;
            box-shadow: 0 2px 8px rgba(0, 0, 0, 0.1);
        }

        textarea {
            width: 100%;
            height: 100px;
            padding: 10px;
            font-size: 1em;
            border-radius: 8px;
            border: 1px solid #ccc;
            resize: vertical;
        }

        button {
            margin-top: 10px;
            padding: 10px 20px;
            background: #007bff;
            border: none;
            color: white;
            border-radius: 8px;
            font-size: 1em;
            cursor: pointer;
        }

        button:disabled {
            background: #999;
        }

        select {
            margin-bottom: 10px;
            padding: 8px;
            border-radius: 8px;
            border: 1px solid #ccc;
            font-size: 1em;
            width: 100%;
        }

        pre {
            white-space: pre-wrap;
            background: #f0f0f0;
            padding: 10px;
            border-radius: 8px;
            min-height: 80px;
        }
    </style>
</head>

<body>
    <h2>🪄 ONNX Paraphraser</h2>
    <label for="modelSelect">Select Model:</label>
    <select id="modelSelect">
        <option value="./model_files/model.onnx">model.onnx</option>
        <option value="./model_files/model_q8.onnx">model_q8.onnx</option>
    </select>

    <p>Enter text to rephrase:</p>
    <textarea id="input" placeholder="Type something here..."></textarea><br>
    <button id="btn">Generate</button>
    <h3>Output:</h3>
    <pre id="output"></pre>

    <script type="module">
        import * as ort from "./wasm/onnxruntime-web.mjs";
        import { AutoTokenizer, env } from "https://cdn.jsdelivr.net/npm/@huggingface/transformers@3.7.6/+esm"

        // Environment setup
        env.localModelPath = './model_files';
        env.allowLocalModels = true;
        env.allowRemoteModels = false;
        env.backends.onnx.wasm.wasmPaths = './wasm/';

        // Globals for caching
        let session = null;
        let tokenizer = null;
        let currentModelPath = null;

        const btn = document.getElementById("btn");
        const inputEl = document.getElementById("input");
        const outputEl = document.getElementById("output");
        const modelSelect = document.getElementById("modelSelect");

        // Utility functions
        function softmax(logits, temperature = 1.0) {
            const adjusted = logits.map(l => l / temperature);
            const max = Math.max(...adjusted);
            const exp = adjusted.map(l => Math.exp(l - max));
            const sum = exp.reduce((a, b) => a + b, 0);
            return exp.map(e => e / sum);
        }

        function topKSample(probabilities, k) {
            const indexed = probabilities.map((p, i) => ({ p, i })).sort((a, b) => b.p - a.p);
            const top = indexed.slice(0, k);
            const sum = top.reduce((a, b) => a + b.p, 0);
            const norm = top.map(t => t.p / sum);
            let r = Math.random(), cum = 0;
            for (let i = 0; i < norm.length; i++) {
                cum += norm[i];
                if (r < cum) return top[i].i;
            }
            return top[0].i;
        }

        // Load model + tokenizer (cached)
        async function loadModelAndTokenizer(modelPath) {
            if (session && tokenizer && currentModelPath === modelPath) {
                console.log("✅ Using cached model:", modelPath);
                return;
            }
            outputEl.textContent = "⏳ Loading model & tokenizer...";
            tokenizer = await AutoTokenizer.from_pretrained("./model_files");
            session = await ort.InferenceSession.create(modelPath, {
                executionProviders: ["wasm"]
            });
            currentModelPath = modelPath;
            outputEl.textContent = `✅ Model loaded: ${modelPath}`;
        }

        btn.onclick = async () => {
            const prompt = inputEl.value.trim();
            const selectedModel = modelSelect.value;

            if (!prompt) {
                outputEl.textContent = "⚠️ Please enter some text.";
                return;
            }

            btn.disabled = true;
            try {
                await loadModelAndTokenizer(selectedModel);

                const endId = (await tokenizer.encode("<|OUTPUT_END|>"))[0];
                let inputText = `\nRephrase the following:\nInput: ${prompt}`;
                const maxTokens = 64, topK = 8, temperature = 0.8;

                outputEl.textContent = "🧠 Running inference...\n\n";
                for (let step = 0; step < maxTokens; step++) {
                    const encoded = await tokenizer(inputText, { return_tensors: "ort" });
                    const feeds = { input_ids: encoded.input_ids.ort_tensor };
                    if (encoded.attention_mask)
                        feeds.attention_mask = encoded.attention_mask.ort_tensor;

                    const outputMap = await session.run(feeds);
                    const logits = outputMap.logits?.data || outputMap[Object.keys(outputMap)[0]].data;
                    const vocabSize = logits.length / encoded.input_ids.data.length;
                    const lastLogits = Array.from(logits.slice(-vocabSize), Number);

                    const probs = softmax(lastLogits, temperature);
                    const nextId = topKSample(probs, topK);
                    const nextToken = await tokenizer.decode([nextId], { skip_special_tokens: false });

                    if (nextToken && !["<|OUTPUT_START|>", "<|OUTPUT_END|>", "Output:"].includes(nextToken.trim())) {
                        outputEl.textContent += nextToken;
                    }

                    inputText += nextToken;
                    if (nextId === endId) break;

                    await new Promise(r => setTimeout(r, 5));
                }

                outputEl.textContent += "\n\n✅ Done";
            } catch (err) {
                console.error(err);
                outputEl.textContent = "❌ Error: " + err.message;
            }

            btn.disabled = false;
        };
    </script>
</body>

</html>