Update Thinking Like Transformers-cn.md
Browse files- Thinking Like Transformers-cn.md +57 -57
Thinking Like Transformers-cn.md
CHANGED
@@ -1,20 +1,20 @@
|
|
1 |
-
# 像Transformer一样思考
|
2 |
-
- [论文](https://arxiv.org/pdf/2106.06981.pdf)来自Gail Weiss, Yoav Goldberg,Eran Yahav.
|
3 |
-
- 博客参考[ Sasha Rush ](https://rush-nlp.com/)和[ Gail Weiss ](https://sgailw.cswp.cs.technion.ac.il/)
|
4 |
-
- 库和交互Notebook:[ srush/raspy ](https://github.com/srush/RASPy)
|
5 |
|
6 |
|
7 |
-
|
|
|
|
|
|
|
|
|
8 |
|
9 |
<center>
|
10 |
|
11 |
-
![](https://
|
12 |
|
13 |
</center>
|
14 |
|
15 |
但是这些图表没有提供任何直观的计算该模型的框架表示。当研究者对于 Transformer 如何工作抱有兴趣时,直观的获取他运行的机制变得十分有用。
|
16 |
|
17 |
-
[像Transformers一样思考](https://arxiv.org/pdf/2106.06981.pdf)提出 transformer 类计算框架。这个框架直接计算和模仿 Transformer 计算。使用[
|
18 |
|
19 |
在这篇博客中,我用 python 复现了 RASP 的变体( RASPy )。该语言大致与原始版本相当,但是多了一些我认为很有趣的变化。通过这些语言,作者 Gail Weiss 的工作,提供了一套具有挑战性的有趣且正确的方式可以帮助了解其工作原理。
|
20 |
|
@@ -33,7 +33,7 @@ flip()
|
|
33 |
```
|
34 |
<center>
|
35 |
|
36 |
-
![](https://
|
37 |
|
38 |
</center>
|
39 |
## 表格目录
|
@@ -53,7 +53,7 @@ flip()
|
|
53 |
|
54 |
<center>
|
55 |
|
56 |
-
![](https://
|
57 |
|
58 |
</center>
|
59 |
|
@@ -65,7 +65,7 @@ tokens
|
|
65 |
|
66 |
<center>
|
67 |
|
68 |
-
![](https://
|
69 |
|
70 |
</center>
|
71 |
|
@@ -77,7 +77,7 @@ tokens.input([5, 2, 4, 5, 2, 2])
|
|
77 |
|
78 |
<center>
|
79 |
|
80 |
-
![](https://
|
81 |
|
82 |
</center>
|
83 |
|
@@ -90,7 +90,7 @@ indices
|
|
90 |
|
91 |
<center>
|
92 |
|
93 |
-
![](https://
|
94 |
|
95 |
</center>
|
96 |
|
@@ -101,7 +101,7 @@ sop.input("goodbye")
|
|
101 |
|
102 |
<center>
|
103 |
|
104 |
-
![](https://
|
105 |
|
106 |
</center>
|
107 |
|
@@ -112,7 +112,7 @@ sop.input("goodbye")
|
|
112 |
|
113 |
<center>
|
114 |
|
115 |
-
![](https://
|
116 |
|
117 |
</center>
|
118 |
|
@@ -124,7 +124,7 @@ tokens == "l"
|
|
124 |
|
125 |
<center>
|
126 |
|
127 |
-
![](https://
|
128 |
|
129 |
</center>
|
130 |
|
@@ -137,7 +137,7 @@ model.input([1, 2, 3, 5, 2])
|
|
137 |
|
138 |
<center>
|
139 |
|
140 |
-
![](https://
|
141 |
|
142 |
</center>
|
143 |
|
@@ -150,7 +150,7 @@ model.input([1, 2, 3, 5, 2])
|
|
150 |
|
151 |
<center>
|
152 |
|
153 |
-
![](https://
|
154 |
|
155 |
</center>
|
156 |
|
@@ -161,7 +161,7 @@ model.input([1, 2, 3, 5, 2])
|
|
161 |
|
162 |
<center>
|
163 |
|
164 |
-
![](
|
165 |
|
166 |
</center>
|
167 |
|
@@ -174,7 +174,7 @@ where((tokens == "h") | (tokens == "l"), tokens, "q")
|
|
174 |
|
175 |
<center>
|
176 |
|
177 |
-
![](https://
|
178 |
|
179 |
</center>
|
180 |
|
@@ -187,7 +187,7 @@ atoi.input("31234")
|
|
187 |
|
188 |
<center>
|
189 |
|
190 |
-
![](https://
|
191 |
|
192 |
</center>
|
193 |
|
@@ -205,7 +205,7 @@ op.input("02-13")
|
|
205 |
|
206 |
<center>
|
207 |
|
208 |
-
![](https://
|
209 |
|
210 |
</center>
|
211 |
|
@@ -214,7 +214,7 @@ op.input("02-13")
|
|
214 |
|
215 |
<center>
|
216 |
|
217 |
-
![](https://
|
218 |
|
219 |
</center>
|
220 |
|
@@ -226,7 +226,7 @@ key(tokens)
|
|
226 |
|
227 |
<center>
|
228 |
|
229 |
-
![](https://
|
230 |
|
231 |
</center>
|
232 |
|
@@ -239,7 +239,7 @@ query(tokens)
|
|
239 |
|
240 |
<center>
|
241 |
|
242 |
-
![](https://
|
243 |
|
244 |
</center>
|
245 |
|
@@ -252,7 +252,7 @@ query(1)
|
|
252 |
|
253 |
<center>
|
254 |
|
255 |
-
![](https://
|
256 |
|
257 |
</center>
|
258 |
|
@@ -266,7 +266,7 @@ eq
|
|
266 |
|
267 |
<center>
|
268 |
|
269 |
-
![](https://
|
270 |
|
271 |
</center>
|
272 |
|
@@ -281,7 +281,7 @@ offset
|
|
281 |
|
282 |
<center>
|
283 |
|
284 |
-
![](https://
|
285 |
|
286 |
</center>
|
287 |
|
@@ -295,7 +295,7 @@ before
|
|
295 |
|
296 |
<center>
|
297 |
|
298 |
-
![](https://
|
299 |
|
300 |
</center>
|
301 |
|
@@ -309,7 +309,7 @@ after
|
|
309 |
|
310 |
<center>
|
311 |
|
312 |
-
![](https://
|
313 |
|
314 |
</center>
|
315 |
|
@@ -321,7 +321,7 @@ before & eq
|
|
321 |
|
322 |
<center>
|
323 |
|
324 |
-
![](https://
|
325 |
|
326 |
</center>
|
327 |
|
@@ -339,7 +339,7 @@ before & eq
|
|
339 |
|
340 |
<center>
|
341 |
|
342 |
-
![](https://
|
343 |
|
344 |
</center>
|
345 |
|
@@ -347,7 +347,7 @@ before & eq
|
|
347 |
|
348 |
<center>
|
349 |
|
350 |
-
![](https://
|
351 |
|
352 |
</center>
|
353 |
|
@@ -362,7 +362,7 @@ length
|
|
362 |
|
363 |
<center>
|
364 |
|
365 |
-
![](https://
|
366 |
|
367 |
</center>
|
368 |
|
@@ -379,7 +379,7 @@ s1
|
|
379 |
|
380 |
<center>
|
381 |
|
382 |
-
![](https://
|
383 |
|
384 |
</center>
|
385 |
|
@@ -391,7 +391,7 @@ s2
|
|
391 |
```
|
392 |
<center>
|
393 |
|
394 |
-
![](https://
|
395 |
|
396 |
</center>
|
397 |
|
@@ -404,7 +404,7 @@ sel
|
|
404 |
|
405 |
<center>
|
406 |
|
407 |
-
![](https://
|
408 |
|
409 |
</center>
|
410 |
|
@@ -417,7 +417,7 @@ sum2.input([1,3,2,2,2])
|
|
417 |
|
418 |
<center>
|
419 |
|
420 |
-
![](https://
|
421 |
|
422 |
</center>
|
423 |
|
@@ -432,17 +432,17 @@ cumsum().input([3, 1, -2, 3, 1])
|
|
432 |
|
433 |
<center>
|
434 |
|
435 |
-
![](https://
|
436 |
|
437 |
</center>
|
438 |
|
439 |
|
440 |
### 层
|
441 |
-
这个语言支持编译更加复杂的 transforms
|
442 |
|
443 |
<center>
|
444 |
|
445 |
-
![](https://
|
446 |
|
447 |
</center>
|
448 |
|
@@ -456,7 +456,7 @@ x.input([3, 2, 3, 5])
|
|
456 |
|
457 |
<center>
|
458 |
|
459 |
-
![](https://
|
460 |
|
461 |
</center>
|
462 |
|
@@ -486,7 +486,7 @@ index(1)
|
|
486 |
|
487 |
<center>
|
488 |
|
489 |
-
![](https://
|
490 |
|
491 |
</center>
|
492 |
|
@@ -504,13 +504,13 @@ shift(2)
|
|
504 |
|
505 |
<center>
|
506 |
|
507 |
-
![](https://
|
508 |
|
509 |
</center>
|
510 |
|
511 |
### 挑战三 :最小化
|
512 |
|
513 |
-
计算序列的最小值。(这一步开始变得困难,我们版本用了2层注意力机制)
|
514 |
|
515 |
```python
|
516 |
def minimum(seq=tokens):
|
@@ -524,7 +524,7 @@ minimum()([5,3,2,5,2])
|
|
524 |
|
525 |
<center>
|
526 |
|
527 |
-
![](https://
|
528 |
|
529 |
</center>
|
530 |
|
@@ -540,7 +540,7 @@ first("l")
|
|
540 |
|
541 |
<center>
|
542 |
|
543 |
-
![](https://
|
544 |
|
545 |
</center>
|
546 |
|
@@ -558,7 +558,7 @@ ralign()("xyz__")
|
|
558 |
|
559 |
<center>
|
560 |
|
561 |
-
![](https://
|
562 |
|
563 |
</center>
|
564 |
|
@@ -581,7 +581,7 @@ split("+", 1)("xyz+zyr")
|
|
581 |
|
582 |
<center>
|
583 |
|
584 |
-
![](https://
|
585 |
|
586 |
</center>
|
587 |
|
@@ -591,7 +591,7 @@ split("+", 0)("xyz+zyr")
|
|
591 |
|
592 |
<center>
|
593 |
|
594 |
-
![](https://
|
595 |
|
596 |
</center>
|
597 |
|
@@ -610,13 +610,13 @@ slide(tokens != "<").input("xxxh<<<l")
|
|
610 |
|
611 |
<center>
|
612 |
|
613 |
-
![](https://
|
614 |
|
615 |
</center>
|
616 |
|
617 |
## 挑战八:增加
|
618 |
|
619 |
-
|
620 |
|
621 |
```python
|
622 |
add().input("683+345")
|
@@ -636,7 +636,7 @@ add().input("683+345")
|
|
636 |
|
637 |
3. 完成加法
|
638 |
|
639 |
-
这些都是1行代码。完整的系统是6个注意力机制。(尽管 Gail
|
640 |
|
641 |
```python
|
642 |
def add(sop=tokens):
|
@@ -652,7 +652,7 @@ add()("683+345")
|
|
652 |
```
|
653 |
<center>
|
654 |
|
655 |
-
![](https://
|
656 |
|
657 |
</center>
|
658 |
|
@@ -663,11 +663,11 @@ add()("683+345")
|
|
663 |
```python
|
664 |
1028
|
665 |
```
|
666 |
-
|
667 |
|
668 |
像 Transformer 和 RASP 语言一样思考。
|
669 |
|
670 |
-
|
671 |
|
672 |
在 https://flann.super.site/ 上查看。
|
673 |
|
@@ -675,5 +675,5 @@ add()("683+345")
|
|
675 |
|
676 |
<hr>
|
677 |
|
678 |
-
|
679 |
-
|
|
|
|
|
|
|
|
|
|
|
1 |
|
2 |
|
3 |
+
- [论文](https://arxiv.org/pdf/2106.06981.pdf) 来自 Gail Weiss, Yoav Goldberg,Eran Yahav.
|
4 |
+
- 博客参考 [Sasha Rush](https://rush-nlp.com/) 和 [Gail Weiss](https://sgailw.cswp.cs.technion.ac.il/)
|
5 |
+
- 库和交互 Notebook: [srush/raspy](https://github.com/srush/RASPy)
|
6 |
+
|
7 |
+
Transformer 模型是 AI 系统的基础。已经有了数不清的关于 "Transformer 如何工作" 的核心结构图表。
|
8 |
|
9 |
<center>
|
10 |
|
11 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/fdf5ede1d0656.jpg)
|
12 |
|
13 |
</center>
|
14 |
|
15 |
但是这些图表没有提供任何直观的计算该模型的框架表示。当研究者对于 Transformer 如何工作抱有兴趣时,直观的获取他运行的机制变得十分有用。
|
16 |
|
17 |
+
[像Transformers一样思考](https://arxiv.org/pdf/2106.06981.pdf)提出 transformer 类计算框架。这个框架直接计算和模仿 Transformer 计算。使用 [RASP](https://github.com/tech-srl/RASP) 编程语言,使每个程序编译成一个特殊的 Transformer 。
|
18 |
|
19 |
在这篇博客中,我用 python 复现了 RASP 的变体( RASPy )。该语言大致与原始版本相当,但是多了一些我认为很有趣的变化。通过这些语言,作者 Gail Weiss 的工作,提供了一套具有挑战性的有趣且正确的方式可以帮助了解其工作原理。
|
20 |
|
|
|
33 |
```
|
34 |
<center>
|
35 |
|
36 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/c2ce067112df6.jpg)
|
37 |
|
38 |
</center>
|
39 |
## 表格目录
|
|
|
53 |
|
54 |
<center>
|
55 |
|
56 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/9b3880c89cddc.jpg)
|
57 |
|
58 |
</center>
|
59 |
|
|
|
65 |
|
66 |
<center>
|
67 |
|
68 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/74668f0bf46e0.jpg)
|
69 |
|
70 |
</center>
|
71 |
|
|
|
77 |
|
78 |
<center>
|
79 |
|
80 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/cae5a2208f808.jpg)
|
81 |
|
82 |
</center>
|
83 |
|
|
|
90 |
|
91 |
<center>
|
92 |
|
93 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/a0a85cb46698a.jpg)
|
94 |
|
95 |
</center>
|
96 |
|
|
|
101 |
|
102 |
<center>
|
103 |
|
104 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/88f599c30ba09.jpg)
|
105 |
|
106 |
</center>
|
107 |
|
|
|
112 |
|
113 |
<center>
|
114 |
|
115 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/20bc37ba063fc.jpg)
|
116 |
|
117 |
</center>
|
118 |
|
|
|
124 |
|
125 |
<center>
|
126 |
|
127 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/2262f13a202af.jpg)
|
128 |
|
129 |
</center>
|
130 |
|
|
|
137 |
|
138 |
<center>
|
139 |
|
140 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/3cf8157521c96.jpg)
|
141 |
|
142 |
</center>
|
143 |
|
|
|
150 |
|
151 |
<center>
|
152 |
|
153 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/9c73afb70fd78.jpg)
|
154 |
|
155 |
</center>
|
156 |
|
|
|
161 |
|
162 |
<center>
|
163 |
|
164 |
+
![](hhttps://devrel.andfun.cn/devrel/posts/2023/01/eefdb8ae5c929.jpg)
|
165 |
|
166 |
</center>
|
167 |
|
|
|
174 |
|
175 |
<center>
|
176 |
|
177 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/46eac496f9afb.jpg)
|
178 |
|
179 |
</center>
|
180 |
|
|
|
187 |
|
188 |
<center>
|
189 |
|
190 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/f7da799e5e2c6.jpg)
|
191 |
|
192 |
</center>
|
193 |
|
|
|
205 |
|
206 |
<center>
|
207 |
|
208 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/888c454a256e6.jpg)
|
209 |
|
210 |
</center>
|
211 |
|
|
|
214 |
|
215 |
<center>
|
216 |
|
217 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/ec55c964b6280.jpg)
|
218 |
|
219 |
</center>
|
220 |
|
|
|
226 |
|
227 |
<center>
|
228 |
|
229 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/d737b1a7b02db.jpg)
|
230 |
|
231 |
</center>
|
232 |
|
|
|
239 |
|
240 |
<center>
|
241 |
|
242 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/e464ca2e26db6.jpg)
|
243 |
|
244 |
</center>
|
245 |
|
|
|
252 |
|
253 |
<center>
|
254 |
|
255 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/eb151bafa0c31.jpg)
|
256 |
|
257 |
</center>
|
258 |
|
|
|
266 |
|
267 |
<center>
|
268 |
|
269 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/f6e54bb8d4e4b.jpg)
|
270 |
|
271 |
</center>
|
272 |
|
|
|
281 |
|
282 |
<center>
|
283 |
|
284 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/cd1c4e50eab36.jpg)
|
285 |
|
286 |
</center>
|
287 |
|
|
|
295 |
|
296 |
<center>
|
297 |
|
298 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/c9084526bd12a.jpg)
|
299 |
|
300 |
</center>
|
301 |
|
|
|
309 |
|
310 |
<center>
|
311 |
|
312 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/f5a8b3e8fe2ff.jpg)
|
313 |
|
314 |
</center>
|
315 |
|
|
|
321 |
|
322 |
<center>
|
323 |
|
324 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/a3d777c4d3bf0.jpg)
|
325 |
|
326 |
</center>
|
327 |
|
|
|
339 |
|
340 |
<center>
|
341 |
|
342 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/2480616d469d3.jpg)
|
343 |
|
344 |
</center>
|
345 |
|
|
|
347 |
|
348 |
<center>
|
349 |
|
350 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/61b054ef9d980.jpg)
|
351 |
|
352 |
</center>
|
353 |
|
|
|
362 |
|
363 |
<center>
|
364 |
|
365 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/fb750cdecc7b5.jpg)
|
366 |
|
367 |
</center>
|
368 |
|
|
|
379 |
|
380 |
<center>
|
381 |
|
382 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/342f3de71068f.jpg)
|
383 |
|
384 |
</center>
|
385 |
|
|
|
391 |
```
|
392 |
<center>
|
393 |
|
394 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/f7d8cd3740935.jpg)
|
395 |
|
396 |
</center>
|
397 |
|
|
|
404 |
|
405 |
<center>
|
406 |
|
407 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/a5840fc2b397e.jpg)
|
408 |
|
409 |
</center>
|
410 |
|
|
|
417 |
|
418 |
<center>
|
419 |
|
420 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/132bb58b8948f.jpg)
|
421 |
|
422 |
</center>
|
423 |
|
|
|
432 |
|
433 |
<center>
|
434 |
|
435 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/bfaace579610a.jpg)
|
436 |
|
437 |
</center>
|
438 |
|
439 |
|
440 |
### 层
|
441 |
+
这个语言支持编译更加复杂的 transforms。他同时通过跟踪每一个运算操作计算层。
|
442 |
|
443 |
<center>
|
444 |
|
445 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/3596032dc760e.jpg)
|
446 |
|
447 |
</center>
|
448 |
|
|
|
456 |
|
457 |
<center>
|
458 |
|
459 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/54d1ded5a9d85.jpg)
|
460 |
|
461 |
</center>
|
462 |
|
|
|
486 |
|
487 |
<center>
|
488 |
|
489 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/f357ce42e5db5.jpg)
|
490 |
|
491 |
</center>
|
492 |
|
|
|
504 |
|
505 |
<center>
|
506 |
|
507 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/152755bcf4f76.jpg)
|
508 |
|
509 |
</center>
|
510 |
|
511 |
### 挑战三 :最小化
|
512 |
|
513 |
+
计算序列的最小值。(这一步开始变得困难,我们版本用了 2 层注意力机制)
|
514 |
|
515 |
```python
|
516 |
def minimum(seq=tokens):
|
|
|
524 |
|
525 |
<center>
|
526 |
|
527 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/f8f4c65d3402d.jpg)
|
528 |
|
529 |
</center>
|
530 |
|
|
|
540 |
|
541 |
<center>
|
542 |
|
543 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/95fd6f19781b3.jpg)
|
544 |
|
545 |
</center>
|
546 |
|
|
|
558 |
|
559 |
<center>
|
560 |
|
561 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/8ec0a6464037a.jpg)
|
562 |
|
563 |
</center>
|
564 |
|
|
|
581 |
|
582 |
<center>
|
583 |
|
584 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/7c4b7272a47b3.jpg)
|
585 |
|
586 |
</center>
|
587 |
|
|
|
591 |
|
592 |
<center>
|
593 |
|
594 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/9f5a2a64f0111.jpg)
|
595 |
|
596 |
</center>
|
597 |
|
|
|
610 |
|
611 |
<center>
|
612 |
|
613 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/27b3d49a00b63.jpg)
|
614 |
|
615 |
</center>
|
616 |
|
617 |
## 挑战八:增加
|
618 |
|
619 |
+
你要执行两个数字的添加。这是步骤。
|
620 |
|
621 |
```python
|
622 |
add().input("683+345")
|
|
|
636 |
|
637 |
3. 完成加法
|
638 |
|
639 |
+
这些都是1行代码。完整的系统是6个注意力机制。(尽管 Gail 说,如果你足够细心则可以在5个中完成!)。
|
640 |
|
641 |
```python
|
642 |
def add(sop=tokens):
|
|
|
652 |
```
|
653 |
<center>
|
654 |
|
655 |
+
![](https://devrel.andfun.cn/devrel/posts/2023/01/0c57f3e6ae5ac.jpg)
|
656 |
|
657 |
</center>
|
658 |
|
|
|
663 |
```python
|
664 |
1028
|
665 |
```
|
666 |
+
非常整洁的东西。如果你对此主题更感兴趣,请务必查看论文:
|
667 |
|
668 |
像 Transformer 和 RASP 语言一样思考。
|
669 |
|
670 |
+
如果你通常对形式语言和神经网络 (FLaNN) 的联系感兴趣,或者认识感兴趣的人,那么这里有一个在线社区,到目前为止非常友好且相当活跃!
|
671 |
|
672 |
在 https://flann.super.site/ 上查看。
|
673 |
|
|
|
675 |
|
676 |
<hr>
|
677 |
|
678 |
+
> 英文原文:[Thinking Like Transformers](https://srush.github.io/raspy/)
|
679 |
+
> 译者:innovation64 (李洋)
|