-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest-encoder-118-unfuse-gelu.html
355 lines (331 loc) · 13.6 KB
/
test-encoder-118-unfuse-gelu.html
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
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta http-equiv="X-UA-Compatible" content="IE=edge" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>ONNXRuntime Web Benchmark Tool</title>
</head>
<style>
body {
font-family: sans-serif;
padding: 20px;
}
h1 {
color: #425066;
font-size: 31px;
margin-top: 0;
}
.loading-stats {
color: #aaa;
font-size: 12px;
margin-top: -12px;
}
.hide {
display: none;
}
.content {
margin-top: 30px;
}
div {
margin-top: 20px;
}
</style>
<body>
<h1>ONNXRuntime Web Benchmark Tool</h1>
<!-- Loading status -->
<div class="loading-stats">Choose options then click 'Run'...</div>
<div>
Number of runs:
<input
type="number"
id="numRuns"
value="1"
min="1"
defaultValue="1"
/>
</div>
<div>
Model:
<select id="model">
<option value="whisper_base_encoder_lm_fp16_layernorm_gelu">whisper_base_encoder_lm_fp16_layernorm_gelu</option>
<option value="whisper_base_decoder_static_non_kvcache_lm_fp16_layernorm_gelu_4dmask">whisper_base_decoder_static_non_kvcache_lm_fp16_layernorm_gelu_4dmask</option>
<option value="whisper_base_decoder_static_kvcache_128_lm_fp16_layernorm_gelu_4dmask">whisper_base_decoder_static_kvcache_128_lm_fp16_layernorm_gelu_4dmask</option>
<option value="whisper_base_decoder_static_non_kvcache_lm_fp16_layernorm_gelu">whisper_base_decoder_static_non_kvcache_lm_fp16_layernorm_gelu_2dmask</option>
<option value="whisper_base_decoder_static_kvcache_128_lm_fp16_layernorm_gelu">whisper_base_decoder_static_kvcache_128_lm_fp16_layernorm_gelu_2dmask</option>
</select>
</div>
<div>
Backend:
<select id="backend">
<option value="webnn">webnn</option>
</select>
</div>
<div>
DeviceType:
<select id="deviceType">
<option value="cpu">WebNN CPU</option>
<option value="gpu">WebNN GPU</option>
<option value="npu" selected>WebNN NPU</option>
</select>
WebNN numThreads:
<input type="number" id="numThreads" defaultValue="0" value="0" />
</div>
<div>
<input type="button" value="Run" id="run" />
</div>
<div id="status" style="font: 1em sans-serif"></div>
<script src="./onnxruntime-web/dist/ort.all.js"></script>
<!-- <script src="https://cdnjs.cloudflare.com/ajax/libs/onnxruntime-web/1.18.0/ort.all.min.js"
integrity="sha512-MHOBvmgua1hQ3KTf3WRm+6UwgeulioSsl38T264/q8+zuL+Z0g0k9+VjvWC3lxAWtENRzUz4Bri9zKsZ/LRZ8w=="
crossorigin="anonymous" referrerpolicy="no-referrer"></script> -->
<script>
function log(i) {
console.log(i);
document.getElementById("status").innerText +=
`\n[${performance.now().toFixed(3)}] ` + i;
}
function generateTensor(dataType, shape, val) {
let size = 1;
shape.forEach((element) => {
size *= element;
});
switch (dataType) {
case "bool":
return new ort.Tensor(
dataType,
Uint8Array.from({ length: size }, () => val),
shape
);
case "uint8":
return new ort.Tensor(
dataType,
Uint8Array.from({ length: size }, () => val),
shape
);
case "uint16":
return new ort.Tensor(
dataType,
Uint16Array.from({ length: size }, () => val),
shape
);
case "float16":
return new ort.Tensor(
dataType,
Uint16Array.from({ length: size }, () => val),
shape
);
case "float32":
return new ort.Tensor(
dataType,
Float32Array.from({ length: size }, () => val),
shape
);
case "int32":
return new ort.Tensor(
dataType,
Int32Array.from({ length: size }, () => val),
shape
);
case "int64":
return new ort.Tensor(
dataType,
BigInt64Array.from({ length: size }, () => val),
shape
);
}
throw new Error(`Input tensor type ${dataType} is unknown`);
}
const type_to_func = {
bool: Uint8Array,
uint8: Uint8Array,
float32: Float32Array,
uint16: Uint16Array,
float16: Uint16Array,
int32: Int32Array,
BigInt64Array: BigInt64Array,
int64: BigInt64Array,
};
function clone(x) {
let feed = {};
for (const [key, value] of Object.entries(x)) {
let func = type_to_func[value.type];
let arrayType = func.from(value.data);
feed[key] = new ort.Tensor(
value.type,
arrayType.slice(0),
value.dims
);
}
return feed;
}
ort.env.wasm.numThreads = 1;
ort.env.wasm.simd = true;
ort.env.wasm.proxy = false;
ort.env.logLevel = "warning"; //"error";
ort.env.debug = false;
// ort.env.wasm.wasmPaths = location.origin + "/onnxruntime-web/ort-web-20240114/dist/";
async function run() {
let feed = {};
const provider = document.getElementById("backend").value;
const modelName = document.getElementById("model").value;
const deviceType = document.getElementById("deviceType").value;
const numThreads = document.getElementById("numThreads").value;
log("entering run ...");
try {
const options = {
executionProviders: [
{
name: provider,
deviceType: deviceType,
powerPreference: "default",
numThreads: parseInt(numThreads),
},
],
//executionProviders: [{name: "webnn", deviceType: 'gpu', powerPreference: 'high-performance' }],
};
const urlPrefix = location.href.includes('github.io') ? 'https://huggingface.co/lwanming/whisper-base-static-shape/resolve/main/' : 'models/';
const modelPath = await getModelOPFS(modelName, `${urlPrefix + modelName}.onnx`, false);
if (modelName.includes('whisper_base_encoder')) {
feed["input_features"] = generateTensor("float16", [1, 80, 3000], 0.5);
} else if (modelName.includes('whisper_base_decoder_static_non_kvcache_lm_fp16')) {
feed['input_ids'] = generateTensor('int32', [1,4], 1);
if (modelName.indexOf('4dmask') != -1) {
feed['attention_mask'] = generateTensor('float16', [1,1,4,4], 1);
} else {
feed['attention_mask'] = generateTensor('int32', [1,4], 1);
}
feed['encoder_hidden_states'] = generateTensor('float16', [1,1500,512], 1);
} else if (modelName.includes('whisper_base_decoder_static_kvcache_128_lm_fp16')) {
feed['input_ids'] = generateTensor('int32', [1,1], 1);
if (modelName.indexOf('4dmask') != -1) {
feed['attention_mask'] = generateTensor('float16', [1,1,1,128], 1);
} else {
feed['attention_mask'] = generateTensor('int64', [1,128], 1n);
}
feed['position_ids'] = generateTensor('int32', [1], 1);
feed['past_key_values.0.decoder.key'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.0.decoder.value'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.0.encoder.key'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.0.encoder.value'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.1.decoder.key'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.1.decoder.value'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.1.encoder.key'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.1.encoder.value'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.2.decoder.key'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.2.decoder.value'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.2.encoder.key'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.2.encoder.value'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.3.decoder.key'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.3.decoder.value'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.3.encoder.key'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.3.encoder.value'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.4.decoder.key'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.4.decoder.value'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.4.encoder.key'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.4.encoder.value'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.5.decoder.key'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.5.decoder.value'] = generateTensor("float16", [1, 8, 127, 64], 0.5);
feed['past_key_values.5.encoder.key'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
feed['past_key_values.5.encoder.value'] = generateTensor("float16", [1, 8, 1500, 64], 0.5);
} else {
throw new Error(`Model ${modelName} is unknown`);
}
options.logSeverityLevel = 0;
log("creating session ...");
console.log("sessionOptions: ", options);
const sess_start = performance.now();
const sess = await ort.InferenceSession.create(modelPath, options);
log(`Session created in ${(performance.now() - sess_start).toFixed(2)} ms`);
log("warmup ...");
console.log("inputs: ", feed);
const output = await sess.run(feed);
console.log("output: ", output);
log("running ...");
let N = document.getElementById("numRuns").value;
N = N === null ? 10 : parseInt(N);
if (N < 1) {
throw new Error("Run Number should be greater than 0!");
}
let inferTimes = [];
for (var i = 0; i < N; i++) {
const start = performance.now();
console.time("ort session.run()");
const outputs = await sess.run(feed);
console.timeEnd("ort session.run()");
//const outputs = await sess.run(feed); // Without clone(), you get DOMException: Failed to execute 'postMessage' on 'Worker': ArrayBuffer at index 0 is already detached.
inferTimes.push(performance.now() - start);
}
// sess.endProfiling();
let intermediateTimings = "";
for (var i = 0; i < N; i++) {
intermediateTimings += `${inferTimes[i].toFixed(2)}, `;
}
log(intermediateTimings);
const totalTime = inferTimes.reduce(
(partialSum, a) => partialSum + a,
0
);
const result = `${modelName}/${provider}, ${(totalTime / N).toFixed(
2
)} ms / iter`;
log(result);
await sess.release();
} catch (e) {
log(e);
}
}
// Get model via Origin Private File System
async function getModelOPFS(name, url, updateModel) {
const root = await navigator.storage.getDirectory();
let fileHandle;
async function updateFile() {
const response = await fetch(url);
const buffer = await readResponse(response);
fileHandle = await root.getFileHandle(name, {create: true});
const writable = await fileHandle.createWritable();
await writable.write(buffer);
await writable.close();
return buffer;
}
if (updateModel) {
return await updateFile();
}
try {
fileHandle = await root.getFileHandle(name);
const blob = await fileHandle.getFile();
return await blob.arrayBuffer();
} catch (e) {
return await updateFile();
}
}
async function readResponse(response) {
const contentLength = response.headers.get('Content-Length');
let total = parseInt(contentLength ?? '0');
let buffer = new Uint8Array(total);
let loaded = 0;
const reader = response.body.getReader();
async function read() {
const {done, value} = await reader.read();
if (done) return;
let newLoaded = loaded + value.length;
if (newLoaded > total) {
total = newLoaded;
let newBuffer = new Uint8Array(total);
newBuffer.set(buffer);
buffer = newBuffer;
}
buffer.set(value, loaded);
loaded = newLoaded;
return read();
}
await read();
return buffer;
}
const runBtn = document.getElementById("run");
runBtn.onclick = async () => {
await run();
};
</script>
</body>
</html>