-
Notifications
You must be signed in to change notification settings - Fork 3
/
index.html
591 lines (530 loc) · 23 KB
/
index.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
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>AI Model Efficiency Toolkit</title>
<meta name="description" content="QuIC presents the AI Model Efficiency Toolkit (AIMET) - a library that provides advanced quantization and compression techniques for trained neural network models.">
<meta name="keywords" content="ai model efficiency toolkit, aimet, qualcomm ai,pruning, quantization, network-compression, automl, deep-neural-networks, network-quantization, model-efficiency, open-source">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1">
<!-- Template Basic Images Start -->
<meta property="og:image" content="path/to/image.jpg">
<link rel="icon" href="img/favicon/favicon.ico">
<link rel="apple-touch-icon" sizes="180x180" href="img/favicon/apple-touch-icon-180x180.png">
<!-- Template Basic Images End -->
<!-- Custom Browsers Color Start -->
<meta name="theme-color" content="#000">
<!-- Custom Browsers Color End -->
<link rel="stylesheet" href="css/main.min.css">
<!-- Load google font
================================================== -->
<script type="text/javascript">
WebFontConfig = {
google: { families: [ 'Catamaran:300,400,600,700', 'Raleway:100,700', 'Roboto:700'] }
};
(function() {
var wf = document.createElement('script');
wf.src = ('https:' == document.location.protocol ? 'https' : 'http') +
'://ajax.googleapis.com/ajax/libs/webfont/1.6.26/webfont.js';
wf.type = 'text/javascript';
wf.async = 'true';
var s = document.getElementsByTagName('script')[0];
s.parentNode.insertBefore(wf, s);
})();
</script>
<!-- Load other scripts
================================================== -->
<script type="text/javascript">
var _html = document.documentElement;
_html.className = _html.className.replace("no-js","js");
</script>
<style>.preloader{width: 100%;height: 100%;position: fixed;background-color: #fff;z-index: 9999;}</style>
</head>
<body>
<div class="preloader"></div>
<header class="header">
<a href="#" class="logo">
<div class="logo__img"></div>
</a>
<ul class="menu">
<li class="menu__item">
<a href="#promo" class="menu__link">Home</a>
</li>
<li class="menu__item">
<a href="#about" class="menu__link">About</a>
</li>
<li class="menu__item">
<a href="#video" class="menu__link">Tutorials</a>
</li>
<li class="menu__item">
<a href="#resources" class="menu__link">Resources</a>
</li>
<li class="menu__item">
<a href="#documentation" class="menu__link">Documentation</a>
</li>
</ul>
<div class="btn-menu">
<div class="one"></div>
<div class="two"></div>
<div class="three"></div>
</div>
</header>
<div class="fixed-menu">
<div class="fixed-menu__header">
<a href="#" class="logo logo--color">
<div class="logo__img"></div>
<div class="logo__title">Qualcoom Innovation Center</div>
</a>
<div class="btn-close">
<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" version="1.1" x="0px" y="0px" viewBox="0 0 47.971 47.971" style="enable-background:new 0 0 47.971 47.971;" xml:space="preserve" width="512px" height="512px">
<path d="M28.228,23.986L47.092,5.122c1.172-1.171,1.172-3.071,0-4.242c-1.172-1.172-3.07-1.172-4.242,0L23.986,19.744L5.121,0.88 c-1.172-1.172-3.07-1.172-4.242,0c-1.172,1.171-1.172,3.071,0,4.242l18.865,18.864L0.879,42.85c-1.172,1.171-1.172,3.071,0,4.242 C1.465,47.677,2.233,47.97,3,47.97s1.535-0.293,2.121-0.879l18.865-18.864L42.85,47.091c0.586,0.586,1.354,0.879,2.121,0.879 s1.535-0.293,2.121-0.879c1.172-1.171,1.172-3.071,0-4.242L28.228,23.986z" fill="#006DF0"/></svg>
</div>
</div>
<div class="fixed-menu__content">
<ul class="mob-menu">
<li class="mob-menu__item">
<a href="#promo" class="mob-menu__link">Home</a>
</li>
<li class="mob-menu__item">
<a href="#about" class="mob-menu__link">About</a>
</li>
<li class="mob-menu__item">
<a href="#video" class="mob-menu__link">Tutorials</a>
</li>
<li class="mob-menu__item">
<a href="#resources" class="mob-menu__link">Resources</a>
</li>
<li class="mob-menu__item">
<a href="#documentation" class="mob-menu__link">Documentation</a>
</li>
</ul>
</div>
</div>
<div class="wrapper">
<img src="imgs/big-triangle-bg-right.png" class="wrapper__bg" alt="">
<section id="promo" class="promo">
<div class="container">
<div class="row align-items-center ">
<div class="col">
<div class="promo__content" data-aos="fade-up">
<h1 class="title title--big title--white promo__title">
AI Model
<span>Efficiency Toolkit (AIMET)</span>
</h1>
<p class="promo__tags">
<span class="tag__img"></span>pruning, quantization, network-compression, automl, deep-neural-networks, network-quantization, model-efficiency, open-source.
</p>
<!-- <form action="#" id="subscribe-top" class="form join-us">
<input name="subscribe_top__email" class="form__input join-us__input" type="email" placeholder="Email">
<button class="form__btn join-us__btn">Join ICO</button>
</form> -->
</div>
<!-- <img src="img/city.png" data-aos="fade-left" alt="" class="promo__img"> -->
</div>
</div>
</div>
<img src="img/aboutBg.png" class="promo__bg" alt="">
</section>
<section id="about" class="about section">
<div class="container">
<div class="row about__first-row">
<div class="col-lg-5" data-aos="fade-right">
<div class="block-header">
<!-- <div class="subtitle subtitle--tire">About AIMET</div> -->
<h2 class="title title--black title--medium">Open-sourcing our
AI Model Efficiency Toolkit</h2>
</div>
<p>
Qualcomm Innovation Center (QuIC) is at the forefront of enabling low-power inference at the edge through its pioneering model-efficiency research. QuIC has a mission to help migrate the ecosystem toward fixed-point inference. With this goal, QuIC presents the AI Model Efficiency Toolkit (AIMET) - a library that provides advanced quantization and compression techniques for trained neural network models. AIMET enables neural networks to run more efficiently on fixed-point AI hardware accelerators.
</p>
<a href="https://github.com/quic/aimet">
<img src="imgs/button_aimet-on-github.png">
</a>
</div>
</div>
</div>
</section>
<section class="process section">
<div class="container">
<img src="imgs/AImodelEfficency.png" class="ai-model-eff-1" alt="">
</div>
</section>
<section id="services" class="service section why">
<div class="container">
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<!-- <div class="subtitle">LOREM IPSUM</div> -->
<h2 class="title title--medium title--black">Why AI Model Efficiency Toolkit?</h2>
</div>
</div>
</div>
<div class="row">
<div class="col-lg-6">
<div class="block-header block-header--animated block-header--left">
<h3 class="title title--small title--black">Performance:</h3>
<p>
Quantized inference is significantly faster than floating point inference. For example, models that we’ve run on the Qualcomm® Hexagon™ DSP rather than on the Qualcomm® Kryo™ CPU have resulted in a 5x to 15x speedup. Plus, an 8-bit model also has a 4x smaller memory footprint relative to a 32-bit model. However, often when quantizing a machine learning model (e.g., from 32-bit floating point to an 8-bit fixed point value), the model accuracy is sacrificed. AIMET solves this problem through novel techniques like data-free quantization that provides state of the art INT8 results as shown in Data-Free Quantization paper ( ICCV’19).
</p>
</div>
</div>
<div class="col-lg-6">
<div class="block-header block-header--animated block-header--left">
<h5 class="title title--small title--black">Scalability:</h5>
<p>
Manual optimization of a neural network for improved efficiency is costly, time-consuming and not scalable with ever increasing AI workloads. AIMET solves this by providing a library that plugs directly into TensorFlow and PyTorch training frameworks for ease of use, allowing developers to call APIs directly from their existing pipelines.
</div>
</div>
</div>
</div>
</section>
<section class="process section">
<div class="container">
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<div class="subtitle">PROCESS</div>
<h2 class="title title--medium title--white">How does it work?</h2>
</div>
</div>
</div>
<img src="imgs/how-does-it-work.png" class="ai-model-eff-1" alt="">
</div>
</section>
<section id="services" class="service section features-API">
<div class="container">
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<!-- <div class="subtitle">LOREM IPSUM</div> -->
<h2 class="title title--medium title--black">Features</h2>
</div>
</div>
</div>
<div class="row">
<div class="col-lg-6">
<div class="block-header block-header--animated block-header--center">
<ul class="about__list-check">
<li class="subtitle">
Quantization
</li>
<li>
Cross-Layer Equalization
<span>Equalize weight tensors to reduce amplitude variation across channels</span>
</li>
<li>
Bias Correction
<span>Corrects shift in layer outputs introduced due to quantization</span>
</li>
<li>
Quantization Simulation
<span>Simulate on-target quantized inference accuracy</span>
</li>
<li>
Fine-tuning
<span>Use quantization sim to train the model further to improve accuracy</span>
</li>
</ul>
</div>
</div>
<div class="col-lg-6">
<div class="block-header block-header--animated block-header--center">
<ul class="about__list-check">
<li class="subtitle">
Compression
</li>
<li>
Spatial SVD
<span>Tensor-decomposition technique to split a large layer into two smaller ones</span>
</li>
<li>
Channel Pruning
<span>Removes redundant input channels from a layer and reconstructs layer weights</span>
</li>
<li>
Automatic selection of per-layer compression ratios
<span>Automatically selects how much to compress each layer in the model</span>
</li>
</ul>
</div>
</div>
<div class="col">
<div class="block-header block-header--animated block-header--center">
<ul class="about__list-check">
<li class="subtitle">
Visualization
</li>
<li>
Visualize weight ranges
</li>
<li>
Visualize per-layer sensitivity to compression
</li>
</ul>
</div>
</div>
</div>
</div>
</section>
<section class="section tables">
<div class="container">
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<div class="subtitle">OUR DATA</div>
<h2 class="title title--medium title--black">What performance benefits can you expect?</h2>
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center tables__text">
<p>
Through a series of simple API calls, AIMET can quantize an existing 32-bit floating-point model to an 8-bit fixed-point model without sacrificing much accuracy and without model fine-tuning. As an example of accuracy maintained, the DFQ method applied to several popular networks, such as MobileNet-v2 and ResNet-50, result in less than 0.9% loss in accuracy all the way down to 8-bit quantization — in an automated way without any training data.
</div>
</div>
</div>
<table class="table">
<tbody>
<tr class="table--header">
<th class="table--round-coner--top-left">Model</th>
<th>FP32 model</th>
<th class="table--round-coner--top-rigth">INT8 model with DFQ</th>
</tr>
<tr>
<td>Mobilenet-v2 (top-1 accuracy)</td>
<td>71.72%</td>
<td>71.08%</td>
</tr>
<tr>
<td>Resinet-50 (top-1 accuracy)</td>
<td>76.05%</td>
<td>75.45%</td>
</tr>
<tr>
<td>Deeplabv3 (mIoU)</td>
<td>72.65%</td>
<td>71.91%</td>
</tr>
<tr class="table--info">
<td colspan="3">Data-free quantization enables INT8 inference with very minimal loss in accuracy relative to the FP32 model.</td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<p>
Through a series of simple API calls, AIMET can also significantly compress models. For popular models, such as Resnet-50 and Resnet-18, compression with spatial SVD plus channel pruning achieves 50% MAC (multiply-accumulate) reduction while retaining accuracy within approximately 1% of the original uncompressed model.
</div>
</div>
</div>
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<table class="table table--last">
<tbody>
<tr class="table--header">
<th class="table--round-coner--top-left">Model <span>(FP32)</span></th>
<th>Uncompressed model</th>
<th class="table--round-coner--top-rigth">Compressed model<span>(50% MAC reduction with SSVD+CP)</span></th>
</tr>
<tr>
<td>Resnet-50 (top-1 accuracy)</td>
<td>76.05%</td>
<td>75.75%</td>
</tr>
<tr>
<td>Resinet-18 (top-1 accuracy)</td>
<td>69.76%</td>
<td>68.56%</td>
</tr>
<tr class="table--info">
<td colspan="3">AIMET compression techniques reduces MACs by 50% while retaining accuracy within approximately 1% of the original model.</td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
</section>
<section id="video" class="news section video-section">
<div class="container">
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<div class="subtitle">TUTORIALS</div>
<h2 class="title title--medium title--black">Watch the videos below to find out how to get the most out of the AI Model Efficiency Toolkit</h2>
</div>
<div class="video-container">
<iframe width="560" height="315" src="https://www.youtube.com/embed/Ia4Qdx0vO8c" frameborder="0"
allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
</div>
<div class="block-header block-header--animated block-header--center">
</div>
<div class="video-container">
<iframe width="560" height="315" src="https://www.youtube.com/embed/SxIMXo3GvWc" frameborder="0"
allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
</div>
<div class="block-header block-header--animated block-header--center">
</div>
<div class="video-container">
<iframe width="560" height="315" src="https://www.youtube.com/embed/zgfy57saqsQ" frameborder="0"
allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
</div>
<div class="block-header block-header--animated block-header--center">
</div>
<div class="video-container">
<iframe width="560" height="315" src="https://www.youtube.com/embed/0-i64DelNkI" frameborder="0"
allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
</div>
<div class="block-header block-header--animated block-header--center">
</div>
<div class="video-container">
<iframe width="560" height="315" src="https://www.youtube.com/embed/QZP2zRb8Vuo" frameborder="0"
allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
</div>
</div>
</div>
<div class="row">
<div class="col-lg-4">
<div class="block-header block-header--animated block-header--center">
</div>
</div>
<div class="col-lg-4">
<div class="block-header block-header--animated block-header--center">
</div>
</div>
<div class="col-lg-4">
<div class="block-header block-header--animated block-header--center">
</div>
</div>
</div>
<div class="row margin">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<div class="subtitle">Model Efficiency Technology</div>
<h2 class="title title--medium title--white">Explore related Qualcomm AI Research papers</h2>
</div>
</div>
</div>
<div class="row">
<div class="col">
<div class="news-carousel owl-carousel">
<a href="http://path.qualcomm.com/Mm0S090uTT5fC00zD0HW0J1" class="news-carousel__item">
<div class="news-carousel__item-subtitle">Webinar</div>
<h3 class="news-carousel__item-title">
Webinar about quantization with Tijmen Blankevoort
</h3>
</a>
<a href="https://arxiv.org/pdf/1906.04721.pdf" class="news-carousel__item">
<div class="news-carousel__item-subtitle">Article</div>
<h3 class="news-carousel__item-title">
Markus Nagel, Mart van Baalen, Tijmen Blankevoort, Max Welling.
</h3>
<p>
Data-Free Quantization Through Weight Equalization and Bias Correction
</p>
<div class="news-carousel__item-data">
IEEE International Conference on Computer Vision (ICCV), Seoul, October 2019 (oral presentation).
</div>
</a>
<a href="https://arxiv.org/abs/2004.10568" class="news-carousel__item">
<div class="news-carousel__item-subtitle">Article</div>
<h3 class="news-carousel__item-title">
Markus Nagel, Rana Ali Amjad, Mart van Baalen, Christos Louizos, Tijmen Blankevoort
</h3>
<p>
Up or Down? Adaptive Rounding for Post-Training Quantization.
</p>
</a>
<a href="https://arxiv.org/abs/1912.09802" class="news-carousel__item">
<div class="news-carousel__item-subtitle">Article</div>
<h3 class="news-carousel__item-title">
Andrey Kuzmin, Markus Nagel, Saurabh Pitre, Sandeep Pendyam, Tijmen Blankevoort, Max Welling.
</h3>
<p>
Taxonomy and Evaluation of Structured Compression of Convolutional Neural Networks.
</p>
</a>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container">
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<div id="resources" class="subtitle">RESOURCES</div>
<h2 class="title title--medium title--black">Check out related blog posts from Qualcomm AI Research:</h2>
</div>
</div>
</div>
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center related-posts">
<ul class="about__list-check">
<li>
<a href="https://www.qualcomm.com/news/onq/2020/05/04/open-sourcing-ai-model-efficiency-toolkit">Introducing AI Model Efficiency Toolkit (blogpost)</a>
</li>
<li>
<a href="https://www.qualcomm.com/news/onq/2019/03/12/heres-why-quantization-matters-ai">Here’s why quantization matters for AI</a>
</li>
<li>
<a href="https://www.qualcomm.com/news/onq/2019/07/30/new-research-quantization-could-revolutionize-power-efficient-ai">New research on quantization could revolutionize power-efficient AI</a>
</li>
<li>
<a href="https://developer.qualcomm.com/software/qualcomm-neural-processing-sdk">Qualcomm® Neural Processing SDK for AI</a>
</li>
</ul>
</div>
</div>
</div>
<div class="row">
<div class="col">
<div class="block-header block-header--animated block-header--center">
<div class="subtitle"></div>
<h2 class="title title--medium title--black">Check out the documentation:</h2>
</div>
</div>
</div>
<div class="row">
<div id="documentation" class="col">
<div class="block-header block-header--animated block-header--center related-posts">
<ul class="about__list-check">
<li>
<a href="releases/latest/user_guide/index.html">AI Model Efficiency Toolkit User Guide</a>
</li>
<li>
<a href="releases/latest/api_docs/index.html">AI Model Efficiency Toolkit API Documentation</a>
</li>
<li>
<a href="https://forums.quicinc.com">AI Model Efficiency Toolkit Forum</a>
</li>
</ul>
</div>
</div>
</div>
</div>
</section>
<footer class="footer ">
<div class="container">
<img class="logo__img-center" src="imgs/footer-icon.png" alt="">
</div>
<div class="col">
<p style="font-size:12px;color:white"> <br><br><br>Qualcomm AI Research is an initiative of Qualcomm Technologies, Inc.<br>
Qualcomm Innovation Center, Inc. is a wholly-owned subsidiary of Qualcomm Technologies, Inc. <br>
AI Model Efficiency Toolkit is a product of Qualcomm Innovation Center, Inc. <br>
Qualcomm Hexagon, Qualcomm Kyro, and Qualcomm Neural Processing SDK are products of Qualcomm Technologies, Inc. and/or its subsidiaries. <br>
</p>
</div>
</footer>
</div>
<script src="https://ajax.googleapis.com/ajax/libs/jquery/2.2.4/jquery.min.js"></script>
<script>window.jQuery || document.write('<script src="js/jquery-2.2.4.min.js"><\/script>')</script>
<script src="js/scripts.min.js"></script>
</body>
</html>