forked from Xilinx/brevitas
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbrevitas.core.quant.html
813 lines (669 loc) · 54.5 KB
/
brevitas.core.quant.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
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
<!DOCTYPE html>
<html class="writer-html5" lang="en" >
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>brevitas.core.quant package — Brevitas 0.2.0b1.dev0+gc32fb31.d20210127 documentation</title>
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<!--[if lt IE 9]>
<script src="_static/js/html5shiv.min.js"></script>
<![endif]-->
<script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<script type="text/javascript" src="_static/js/theme.js"></script>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search" >
<a href="index.html" class="icon icon-home"> Brevitas
</a>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<!-- Local TOC -->
<div class="local-toc"><ul>
<li><a class="reference internal" href="#">brevitas.core.quant package</a><ul>
<li><a class="reference internal" href="#submodules">Submodules</a></li>
<li><a class="reference internal" href="#module-brevitas.core.quant.binary">brevitas.core.quant.binary module</a></li>
<li><a class="reference internal" href="#module-brevitas.core.quant.delay">brevitas.core.quant.delay module</a></li>
<li><a class="reference internal" href="#module-brevitas.core.quant.int">brevitas.core.quant.int module</a></li>
<li><a class="reference internal" href="#module-brevitas.core.quant.int_base">brevitas.core.quant.int_base module</a></li>
<li><a class="reference internal" href="#module-brevitas.core.quant.ternary">brevitas.core.quant.ternary module</a></li>
<li><a class="reference internal" href="#module-brevitas.core.quant">Module contents</a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
<nav class="wy-nav-top" aria-label="top navigation">
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="index.html">Brevitas</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="wy-breadcrumbs">
<li><a href="index.html" class="icon icon-home"></a> »</li>
<li>brevitas.core.quant package</li>
<li class="wy-breadcrumbs-aside">
<a href="_sources/brevitas.core.quant.rst.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<div class="section" id="brevitas-core-quant-package">
<h1>brevitas.core.quant package<a class="headerlink" href="#brevitas-core-quant-package" title="Permalink to this headline">¶</a></h1>
<div class="section" id="submodules">
<h2>Submodules<a class="headerlink" href="#submodules" title="Permalink to this headline">¶</a></h2>
</div>
<div class="section" id="module-brevitas.core.quant.binary">
<span id="brevitas-core-quant-binary-module"></span><h2>brevitas.core.quant.binary module<a class="headerlink" href="#module-brevitas.core.quant.binary" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="brevitas.core.quant.binary.BinaryQuant">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.binary.</code><code class="sig-name descname">BinaryQuant</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">scaling_impl</span></em>, <em class="sig-param"><span class="n">quant_delay_steps</span><span class="o">=</span><span class="default_value">0</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.binary.BinaryQuant" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<p>ScriptModule that implements scaled uniform binary quantization of an input tensor.
Quantization is performed with <a class="reference internal" href="brevitas.function.html#brevitas.function.ops_ste.binary_sign_ste" title="brevitas.function.ops_ste.binary_sign_ste"><code class="xref py py-func docutils literal notranslate"><span class="pre">binary_sign_ste()</span></code></a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>scaling_impl</strong> (<em>Module</em>) – Module that returns a scale factor.</p></li>
<li><p><strong>quant_delay_steps</strong> (<em>int</em>) – Number of training steps to delay quantization for. Default: 0</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><dl class="simple">
<dt>Quantized output in de-quantized format, scale,</dt><dd><p>zero-point, bit_width.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>Tuple[Tensor, Tensor, Tensor, Tensor]</p>
</dd>
</dl>
<p class="rubric">Examples</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.scaling</span> <span class="kn">import</span> <span class="n">ConstScaling</span>
<span class="gp">>>> </span><span class="n">binary_quant</span> <span class="o">=</span> <span class="n">BinaryQuant</span><span class="p">(</span><span class="n">ConstScaling</span><span class="p">(</span><span class="mf">0.1</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">([</span><span class="mf">0.04</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.6</span><span class="p">,</span> <span class="mf">3.3</span><span class="p">])</span>
<span class="gp">>>> </span><span class="n">out</span><span class="p">,</span> <span class="n">scale</span><span class="p">,</span> <span class="n">zero_point</span><span class="p">,</span> <span class="n">bit_width</span> <span class="o">=</span> <span class="n">binary_quant</span><span class="p">(</span><span class="n">inp</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">out</span>
<span class="go">tensor([ 0.1000, -0.1000, 0.1000])</span>
<span class="gp">>>> </span><span class="n">scale</span>
<span class="go">tensor(0.1000)</span>
<span class="gp">>>> </span><span class="n">zero_point</span>
<span class="go">tensor(0.)</span>
<span class="gp">>>> </span><span class="n">bit_width</span>
<span class="go">tensor(1.)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<dl class="simple">
<dt>Maps to quant_type == QuantType.BINARY == ‘BINARY’ == ‘binary’ when applied to weights</dt><dd><p>in higher-level APIs.</p>
</dd>
</dl>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Set env variable BREVITAS_JIT=1 to enable TorchScript compilation of this module.</p>
</div>
<dl class="py method">
<dt id="brevitas.core.quant.binary.BinaryQuant.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.binary.BinaryQuant.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-data docutils literal notranslate"><span class="pre">Tuple</span></code>[<code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>]</p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.binary.BinaryQuant.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.binary.BinaryQuant.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
<dl class="py class">
<dt id="brevitas.core.quant.binary.ClampedBinaryQuant">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.binary.</code><code class="sig-name descname">ClampedBinaryQuant</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">scaling_impl</span></em>, <em class="sig-param"><span class="n">quant_delay_steps</span><span class="o">=</span><span class="default_value">0</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.binary.ClampedBinaryQuant" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<p>ScriptModule that implements scaled uniform binary quantization of an input tensor. Before
going through quantization, the input tensor is clamped between (- scale, scale), which
on the backward pass zeroes gradients corresponding to inputs outside that range.
Quantization is performed with <a class="reference internal" href="brevitas.function.html#brevitas.function.ops_ste.binary_sign_ste" title="brevitas.function.ops_ste.binary_sign_ste"><code class="xref py py-func docutils literal notranslate"><span class="pre">binary_sign_ste()</span></code></a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>scaling_impl</strong> (<em>Module</em>) – Module that returns a scale factor.</p></li>
<li><p><strong>quant_delay_steps</strong> (<em>int</em>) – Number of training steps to delay quantization for. Default: 0</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><dl class="simple">
<dt>Quantized output in de-quantized format, scale,</dt><dd><p>zero-point, bit_width.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>Tuple[Tensor, Tensor, Tensor, Tensor]</p>
</dd>
</dl>
<p class="rubric">Examples</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.scaling</span> <span class="kn">import</span> <span class="n">ConstScaling</span>
<span class="gp">>>> </span><span class="n">binary_quant</span> <span class="o">=</span> <span class="n">ClampedBinaryQuant</span><span class="p">(</span><span class="n">ConstScaling</span><span class="p">(</span><span class="mf">0.1</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">([</span><span class="mf">0.04</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.6</span><span class="p">,</span> <span class="mf">3.3</span><span class="p">])</span><span class="o">.</span><span class="n">requires_grad_</span><span class="p">(</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">out</span><span class="p">,</span> <span class="n">scale</span><span class="p">,</span> <span class="n">zero_point</span><span class="p">,</span> <span class="n">bit_width</span> <span class="o">=</span> <span class="n">binary_quant</span><span class="p">(</span><span class="n">inp</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">out</span>
<span class="go">tensor([ 0.1000, -0.1000, 0.1000], grad_fn=<MulBackward0>)</span>
<span class="gp">>>> </span><span class="n">out</span><span class="o">.</span><span class="n">backward</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">([</span><span class="mf">1.0</span><span class="p">,</span> <span class="mf">1.0</span><span class="p">,</span> <span class="mf">1.0</span><span class="p">]))</span>
<span class="gp">>>> </span><span class="n">inp</span><span class="o">.</span><span class="n">grad</span>
<span class="go">tensor([0.1000, 0.0000, 0.0000])</span>
<span class="gp">>>> </span><span class="n">scale</span>
<span class="go">tensor(0.1000)</span>
<span class="gp">>>> </span><span class="n">zero_point</span>
<span class="go">tensor(0.)</span>
<span class="gp">>>> </span><span class="n">bit_width</span>
<span class="go">tensor(1.)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<dl class="simple">
<dt>Maps to quant_type == QuantType.BINARY == ‘BINARY’ == ‘binary’ when applied to activations</dt><dd><p>in higher-level APIs.</p>
</dd>
</dl>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Set env variable BREVITAS_JIT=1 to enable TorchScript compilation of this module.</p>
</div>
<dl class="py method">
<dt id="brevitas.core.quant.binary.ClampedBinaryQuant.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.binary.ClampedBinaryQuant.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-data docutils literal notranslate"><span class="pre">Tuple</span></code>[<code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>]</p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.binary.ClampedBinaryQuant.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.binary.ClampedBinaryQuant.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
</div>
<div class="section" id="module-brevitas.core.quant.delay">
<span id="brevitas-core-quant-delay-module"></span><h2>brevitas.core.quant.delay module<a class="headerlink" href="#module-brevitas.core.quant.delay" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="brevitas.core.quant.delay.DelayWrapper">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.delay.</code><code class="sig-name descname">DelayWrapper</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">quant_delay_steps</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.delay.DelayWrapper" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<dl class="py method">
<dt id="brevitas.core.quant.delay.DelayWrapper.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em>, <em class="sig-param"><span class="n">y</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.delay.DelayWrapper.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.delay.DelayWrapper.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.delay.DelayWrapper.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
</div>
<div class="section" id="module-brevitas.core.quant.int">
<span id="brevitas-core-quant-int-module"></span><h2>brevitas.core.quant.int module<a class="headerlink" href="#module-brevitas.core.quant.int" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="brevitas.core.quant.int.PrescaledRestrictIntQuant">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.int.</code><code class="sig-name descname">PrescaledRestrictIntQuant</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">int_quant</span></em>, <em class="sig-param"><span class="n">bit_width_impl</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int.PrescaledRestrictIntQuant" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<dl class="py method">
<dt id="brevitas.core.quant.int.PrescaledRestrictIntQuant.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em>, <em class="sig-param"><span class="n">scale</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int.PrescaledRestrictIntQuant.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-data docutils literal notranslate"><span class="pre">Tuple</span></code>[<code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>]</p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.int.PrescaledRestrictIntQuant.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.int.PrescaledRestrictIntQuant.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
<dl class="py class">
<dt id="brevitas.core.quant.int.PrescaledRestrictIntQuantWithInputBitWidth">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.int.</code><code class="sig-name descname">PrescaledRestrictIntQuantWithInputBitWidth</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">int_quant</span></em>, <em class="sig-param"><span class="n">bit_width_impl</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int.PrescaledRestrictIntQuantWithInputBitWidth" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<p>ScriptModule that wraps around an integer quantization implementation like
<code class="xref py py-class docutils literal notranslate"><span class="pre">IntQuant</span></code>. Zero-point is set to zero, scale is taken as input,
bit-width is computed from an input bit-width.</p>
<blockquote>
<div><dl>
<dt>Args:</dt><dd><p>int_quant (Module): Module that implements integer quantization.
bit_width_impl (Module): Module that takes the input bit-width in and returns the bit-width</p>
<blockquote>
<div><p>to be used for quantization.</p>
</div></blockquote>
</dd>
</dl>
</div></blockquote>
<dl class="field-list simple">
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p><dl class="simple">
<dt>Quantized output in de-quantized format, scale,</dt><dd><p>zero-point, bit_width.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-even">Return type</dt>
<dd class="field-even"><p>Tuple[Tensor, Tensor, Tensor, Tensor]</p>
</dd>
</dl>
<p class="rubric">Examples</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.scaling</span> <span class="kn">import</span> <span class="n">ConstScaling</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.function_wrapper</span> <span class="kn">import</span> <span class="n">Identity</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.quant</span> <span class="kn">import</span> <span class="n">IntQuant</span>
<span class="gp">>>> </span><span class="n">int_quant</span> <span class="o">=</span> <span class="n">IntQuant</span><span class="p">(</span><span class="n">narrow_range</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">signed</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">int_quant_wrapper</span> <span class="o">=</span> <span class="n">PrescaledRestrictIntQuantWithInputBitWidth</span><span class="p">(</span><span class="n">int_quant</span><span class="p">,</span> <span class="n">Identity</span><span class="p">())</span>
<span class="gp">>>> </span><span class="n">scale</span><span class="p">,</span> <span class="n">input_bit_width</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="mf">0.01</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="mf">4.</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">([</span><span class="mf">0.042</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.053</span><span class="p">,</span> <span class="mf">0.31</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.44</span><span class="p">])</span>
<span class="gp">>>> </span><span class="n">out</span><span class="p">,</span> <span class="n">scale</span><span class="p">,</span> <span class="n">zero_point</span><span class="p">,</span> <span class="n">bit_width</span> <span class="o">=</span> <span class="n">int_quant_wrapper</span><span class="p">(</span><span class="n">inp</span><span class="p">,</span> <span class="n">scale</span><span class="p">,</span> <span class="n">input_bit_width</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">out</span>
<span class="go">tensor([ 0.0400, -0.0500, 0.0700, -0.0700])</span>
<span class="gp">>>> </span><span class="n">scale</span>
<span class="go">tensor(0.0100)</span>
<span class="gp">>>> </span><span class="n">zero_point</span>
<span class="go">tensor(0.)</span>
<span class="gp">>>> </span><span class="n">bit_width</span>
<span class="go">tensor(4.)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Set env variable BREVITAS_JIT=1 to enable TorchScript compilation of this module.</p>
</div>
<dl class="py method">
<dt id="brevitas.core.quant.int.PrescaledRestrictIntQuantWithInputBitWidth.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em>, <em class="sig-param"><span class="n">scale</span></em>, <em class="sig-param"><span class="n">input_bit_width</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int.PrescaledRestrictIntQuantWithInputBitWidth.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-data docutils literal notranslate"><span class="pre">Tuple</span></code>[<code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>]</p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.int.PrescaledRestrictIntQuantWithInputBitWidth.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.int.PrescaledRestrictIntQuantWithInputBitWidth.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
<dl class="py class">
<dt id="brevitas.core.quant.int.RescalingIntQuant">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.int.</code><code class="sig-name descname">RescalingIntQuant</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">int_quant</span></em>, <em class="sig-param"><span class="n">scaling_impl</span></em>, <em class="sig-param"><span class="n">int_scaling_impl</span></em>, <em class="sig-param"><span class="n">zero_point_impl</span></em>, <em class="sig-param"><span class="n">bit_width_impl</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int.RescalingIntQuant" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<p>ScriptModule that wraps around an integer quantization implementation like
<code class="xref py py-class docutils literal notranslate"><span class="pre">IntQuant</span></code>. Scale, zero-point and bit-width are returned from their
respective implementations and passed on to the integer quantization implementation.</p>
<blockquote>
<div><dl>
<dt>Args:</dt><dd><p>int_quant (Module): Module that implements integer quantization.
scaling_impl (Module): Module that takes in the input to quantize and returns a scale factor,</p>
<blockquote>
<div><p>here interpreted as threshold on the floating-point range of quantization.</p>
</div></blockquote>
<dl class="simple">
<dt>int_scaling_impl (Module): Module that takes in a bit-width and returns an integer scale</dt><dd><p>factor, here interpreted as threshold on the integer range of quantization.</p>
</dd>
</dl>
<p>zero_point_impl (Module): Module that returns an integer zero-point.
bit_width_impl (Module): Module that returns a bit-width.</p>
</dd>
</dl>
</div></blockquote>
<dl class="field-list simple">
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p><dl class="simple">
<dt>Quantized output in de-quantized format, scale,</dt><dd><p>zero-point, bit_width.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-even">Return type</dt>
<dd class="field-even"><p>Tuple[Tensor, Tensor, Tensor, Tensor]</p>
</dd>
</dl>
<p class="rubric">Examples</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.scaling</span> <span class="kn">import</span> <span class="n">ConstScaling</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.zero_point</span> <span class="kn">import</span> <span class="n">ZeroZeroPoint</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.scaling</span> <span class="kn">import</span> <span class="n">IntScaling</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.quant</span> <span class="kn">import</span> <span class="n">IntQuant</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.bit_width</span> <span class="kn">import</span> <span class="n">BitWidthConst</span>
<span class="gp">>>> </span><span class="n">int_quant_wrapper</span> <span class="o">=</span> <span class="n">RescalingIntQuant</span><span class="p">(</span>
<span class="gp">... </span> <span class="n">IntQuant</span><span class="p">(</span><span class="n">narrow_range</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">signed</span><span class="o">=</span><span class="kc">True</span><span class="p">),</span>
<span class="gp">... </span> <span class="n">ConstScaling</span><span class="p">(</span><span class="mf">0.1</span><span class="p">),</span>
<span class="gp">... </span> <span class="n">IntScaling</span><span class="p">(</span><span class="n">signed</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">narrow_range</span><span class="o">=</span><span class="kc">True</span><span class="p">),</span>
<span class="gp">... </span> <span class="n">ZeroZeroPoint</span><span class="p">(),</span>
<span class="gp">... </span> <span class="n">BitWidthConst</span><span class="p">(</span><span class="mi">4</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">([</span><span class="mf">0.042</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.053</span><span class="p">,</span> <span class="mf">0.31</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.44</span><span class="p">])</span>
<span class="gp">>>> </span><span class="n">out</span><span class="p">,</span> <span class="n">scale</span><span class="p">,</span> <span class="n">zero_point</span><span class="p">,</span> <span class="n">bit_width</span> <span class="o">=</span> <span class="n">int_quant_wrapper</span><span class="p">(</span><span class="n">inp</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">out</span>
<span class="go">tensor([ 0.0429, -0.0571, 0.1000, -0.1000])</span>
<span class="gp">>>> </span><span class="n">scale</span>
<span class="go">tensor(0.0143)</span>
<span class="gp">>>> </span><span class="n">zero_point</span>
<span class="go">tensor(0.)</span>
<span class="gp">>>> </span><span class="n">bit_width</span>
<span class="go">tensor(4.)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>scale = scaling_impl(x) / int_scaling_impl(bit_width)</p>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Set env variable BREVITAS_JIT=1 to enable TorchScript compilation of this module.</p>
</div>
<dl class="py method">
<dt id="brevitas.core.quant.int.RescalingIntQuant.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int.RescalingIntQuant.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-data docutils literal notranslate"><span class="pre">Tuple</span></code>[<code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>]</p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.int.RescalingIntQuant.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.int.RescalingIntQuant.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
<dl class="py class">
<dt id="brevitas.core.quant.int.TruncIntQuant">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.int.</code><code class="sig-name descname">TruncIntQuant</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">float_to_int_impl</span></em>, <em class="sig-param"><span class="n">bit_width_impl</span></em>, <em class="sig-param"><span class="n">quant_delay_steps</span><span class="o">=</span><span class="default_value">None</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int.TruncIntQuant" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<dl class="py method">
<dt id="brevitas.core.quant.int.TruncIntQuant.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em>, <em class="sig-param"><span class="n">scale</span></em>, <em class="sig-param"><span class="n">zero_point</span></em>, <em class="sig-param"><span class="n">input_bit_width</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int.TruncIntQuant.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-data docutils literal notranslate"><span class="pre">Tuple</span></code>[<code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>]</p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.int.TruncIntQuant.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.int.TruncIntQuant.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
</div>
<div class="section" id="module-brevitas.core.quant.int_base">
<span id="brevitas-core-quant-int-base-module"></span><h2>brevitas.core.quant.int_base module<a class="headerlink" href="#module-brevitas.core.quant.int_base" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="brevitas.core.quant.int_base.IntQuant">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.int_base.</code><code class="sig-name descname">IntQuant</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">narrow_range</span></em>, <em class="sig-param"><span class="n">signed</span></em>, <em class="sig-param"><span class="n">float_to_int_impl</span><span class="o">=</span><span class="default_value">RoundSte()</span></em>, <em class="sig-param"><span class="n">tensor_clamp_impl</span><span class="o">=</span><span class="default_value">TensorClamp()</span></em>, <em class="sig-param"><span class="n">quant_delay_steps</span><span class="o">=</span><span class="default_value">0</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int_base.IntQuant" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<p>ScriptModule that implements scale, shifted, uniform integer quantization of an input tensor,
according to an input scale, zero-point and bit-width.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>narrow_range</strong> (<em>bool</em>) – Flag that determines whether restrict quantization to a narrow range or not.</p></li>
<li><p><strong>signed</strong> (<em>bool</em>) – Flag that determines whether to quantize to a signed range or not.</p></li>
<li><p><strong>float_to_int_impl</strong> (<em>Module</em>) – Module that performs the conversion from floating point to
integer representation. Default: RoundSte()</p></li>
<li><p><strong>tensor_clamp_impl</strong> (<em>Module</em>) – Module that performs clamping. Default: TensorClamp()</p></li>
<li><p><strong>quant_delay_steps</strong> (<em>int</em>) – Number of training steps to delay quantization for. Default: 0</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>Quantized output in de-quantized format.</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>Tensor</p>
</dd>
</dl>
<p class="rubric">Examples</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.scaling</span> <span class="kn">import</span> <span class="n">ConstScaling</span>
<span class="gp">>>> </span><span class="n">int_quant</span> <span class="o">=</span> <span class="n">IntQuant</span><span class="p">(</span><span class="n">narrow_range</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">signed</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">scale</span><span class="p">,</span> <span class="n">zero_point</span><span class="p">,</span> <span class="n">bit_width</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="mf">0.01</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="mf">0.</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="mf">4.</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">([</span><span class="mf">0.042</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.053</span><span class="p">,</span> <span class="mf">0.31</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.44</span><span class="p">])</span>
<span class="gp">>>> </span><span class="n">out</span> <span class="o">=</span> <span class="n">int_quant</span><span class="p">(</span><span class="n">scale</span><span class="p">,</span> <span class="n">zero_point</span><span class="p">,</span> <span class="n">bit_width</span><span class="p">,</span> <span class="n">inp</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">out</span>
<span class="go">tensor([ 0.0400, -0.0500, 0.0700, -0.0700])</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Maps to quant_type == QuantType.INT == ‘INT’ == ‘int’ in higher-level APIs.</p>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Set env variable BREVITAS_JIT=1 to enable TorchScript compilation of this module.</p>
</div>
<dl class="py method">
<dt id="brevitas.core.quant.int_base.IntQuant.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">scale</span></em>, <em class="sig-param"><span class="n">zero_point</span></em>, <em class="sig-param"><span class="n">bit_width</span></em>, <em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int_base.IntQuant.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></p>
</dd>
</dl>
</dd></dl>
<dl class="py method">
<dt id="brevitas.core.quant.int_base.IntQuant.max_int">
<code class="sig-name descname">max_int</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">bit_width</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int_base.IntQuant.max_int" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
<dl class="py method">
<dt id="brevitas.core.quant.int_base.IntQuant.min_int">
<code class="sig-name descname">min_int</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">bit_width</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int_base.IntQuant.min_int" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
<dl class="py method">
<dt id="brevitas.core.quant.int_base.IntQuant.to_int">
<code class="sig-name descname">to_int</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">scale</span></em>, <em class="sig-param"><span class="n">zero_point</span></em>, <em class="sig-param"><span class="n">bit_width</span></em>, <em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.int_base.IntQuant.to_int" title="Permalink to this definition">¶</a></dt>
<dd><dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code></p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.int_base.IntQuant.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.int_base.IntQuant.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
</div>
<div class="section" id="module-brevitas.core.quant.ternary">
<span id="brevitas-core-quant-ternary-module"></span><h2>brevitas.core.quant.ternary module<a class="headerlink" href="#module-brevitas.core.quant.ternary" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="brevitas.core.quant.ternary.TernaryQuant">
<em class="property">class </em><code class="sig-prename descclassname">brevitas.core.quant.ternary.</code><code class="sig-name descname">TernaryQuant</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">scaling_impl</span></em>, <em class="sig-param"><span class="n">threshold</span></em>, <em class="sig-param"><span class="n">quant_delay_steps</span><span class="o">=</span><span class="default_value">None</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.ternary.TernaryQuant" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
<p>ScriptModule that implements scaled uniform ternary quantization of an input tensor.
Quantization is performed with <a class="reference internal" href="brevitas.function.html#brevitas.function.ops_ste.ternary_sign_ste" title="brevitas.function.ops_ste.ternary_sign_ste"><code class="xref py py-func docutils literal notranslate"><span class="pre">ternary_sign_ste()</span></code></a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>scaling_impl</strong> (<em>Module</em>) – Module that returns a scale factor.</p></li>
<li><p><strong>threshold</strong> (<em>float</em>) – Ternarization threshold w.r.t. to the scale factor.</p></li>
<li><p><strong>quant_delay_steps</strong> (<em>int</em>) – Number of training steps to delay quantization for. Default: 0</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><dl class="simple">
<dt>Quantized output in de-quantized format, scale,</dt><dd><p>zero-point, bit_width.</p>
</dd>
</dl>
</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>Tuple[Tensor, Tensor, Tensor, Tensor]</p>
</dd>
</dl>
<p class="rubric">Examples</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">brevitas.core.scaling</span> <span class="kn">import</span> <span class="n">ConstScaling</span>
<span class="gp">>>> </span><span class="n">ternary_quant</span> <span class="o">=</span> <span class="n">TernaryQuant</span><span class="p">(</span><span class="n">ConstScaling</span><span class="p">(</span><span class="mf">1.0</span><span class="p">),</span> <span class="mf">0.5</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">([</span><span class="mf">0.04</span><span class="p">,</span> <span class="o">-</span><span class="mf">0.6</span><span class="p">,</span> <span class="mf">3.3</span><span class="p">])</span>
<span class="gp">>>> </span><span class="n">out</span><span class="p">,</span> <span class="n">scale</span><span class="p">,</span> <span class="n">zero_point</span><span class="p">,</span> <span class="n">bit_width</span> <span class="o">=</span> <span class="n">ternary_quant</span><span class="p">(</span><span class="n">inp</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">out</span>
<span class="go">tensor([ 0., -1., 1.])</span>
<span class="gp">>>> </span><span class="n">scale</span>
<span class="go">tensor(1.)</span>
<span class="gp">>>> </span><span class="n">zero_point</span>
<span class="go">tensor(0.)</span>
<span class="gp">>>> </span><span class="n">bit_width</span>
<span class="go">tensor(2.)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Maps to quant_type == QuantType.TERNARY == ‘TERNARY’ == ‘ternary’ in higher-level APIs.</p>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Set env variable BREVITAS_JIT=1 to enable TorchScript compilation of this module.</p>
</div>
<dl class="py method">
<dt id="brevitas.core.quant.ternary.TernaryQuant.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#brevitas.core.quant.ternary.TernaryQuant.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><code class="xref py py-data docutils literal notranslate"><span class="pre">Tuple</span></code>[<code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>, <code class="xref py py-class docutils literal notranslate"><span class="pre">Tensor</span></code>]</p>
</dd>
</dl>
</dd></dl>
<dl class="py attribute">
<dt id="brevitas.core.quant.ternary.TernaryQuant.training">
<code class="sig-name descname">training</code><em class="property">: bool</em><a class="headerlink" href="#brevitas.core.quant.ternary.TernaryQuant.training" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</dd></dl>
</div>
<div class="section" id="module-brevitas.core.quant">
<span id="module-contents"></span><h2>Module contents<a class="headerlink" href="#module-brevitas.core.quant" title="Permalink to this headline">¶</a></h2>
</div>
</div>
</div>
</div>
<footer>
<hr/>
<div role="contentinfo">
<p>
© Copyright 2021 - Xilinx, Inc.
</p>
</div>
Built with <a href="https://www.sphinx-doc.org/">Sphinx</a> using a
<a href="https://github.com/readthedocs/sphinx_rtd_theme">theme</a>
provided by <a href="https://readthedocs.org">Read the Docs</a>.
</footer>
</div>
</div>
</section>
</div>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
</body>
</html>