/
textsummarize.Rmd
287 lines (237 loc) · 14.8 KB
/
textsummarize.Rmd
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
---
title: "Text-summarization"
output:
rmarkdown::html_vignette
vignette: >
%\VignetteIndexEntry{Text-summarization}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
---
```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE, eval = FALSE)
```
## Intro
First, we need to install ```blurr module``` for [Transformers](https://github.com/huggingface/transformers) integration.
```{r}
reticulate::py_install('ohmeow-blurr',pip = TRUE)
```
## Dataset
Get dataset from the link:
```{r}
library(fastai)
library(magrittr)
library(zeallot)
cnndm_df = data.table::fread('https://raw.githubusercontent.com/ohmeow/blurr/master/nbs/cnndm_sample.csv')
```
## Pretrained model
Get pretrained BART model from ```transformers```:
```{r}
transformers = transformers()
BartForConditionalGeneration = transformers$BartForConditionalGeneration
pretrained_model_name = "facebook/bart-large-cnn"
c(hf_arch, hf_config, hf_tokenizer, hf_model) %<-%
get_hf_objects(pretrained_model_name,model_cls=BartForConditionalGeneration)
```
## Datablock
Create datablock and see batch:
```{r}
before_batch_tfm = HF_SummarizationBeforeBatchTransform(hf_arch,
hf_tokenizer, max_length=c(256, 130))
blocks = list(HF_Text2TextBlock(before_batch_tfms=before_batch_tfm,
input_return_type=HF_SummarizationInput), noop())
dblock = DataBlock(blocks=blocks,
get_x=ColReader('article'),
get_y=ColReader('highlights'),
splitter=RandomSplitter())
dls = dblock %>% dataloaders(cnndm_df, bs=2)
dls %>% one_batch()
```
```
[[1]]
[[1]]$input_ids
tensor([[ 0, 36, 16256, 43, 480, 2193, 7, 62, 7, 158,
135, 9, 70, 684, 4707, 6, 1625, 16, 4984, 25,
65, 9, 5, 144, 39865, 32273, 3806, 15, 5, 5518,
4, 20, 9544, 3455, 9, 2147, 464, 8, 1050, 29779,
26284, 15, 1632, 11534, 32, 6, 959, 6, 5608, 5,
8066, 9, 5, 247, 18, 4066, 7892, 4, 178, 89,
16, 10, 372, 432, 7, 2217, 4, 96, 5, 315,
3076, 9356, 4928, 36, 4154, 9662, 43, 623, 12978, 23853,
2521, 18, 889, 9, 10721, 625, 32273, 749, 1625, 5546,
365, 212, 4, 20, 889, 3372, 10, 333, 9, 601,
749, 14, 19655, 5, 1647, 9, 5, 3875, 18, 4707,
8, 32, 3891, 1687, 2778, 39865, 32273, 4, 1740, 63,
23491, 28919, 11, 5, 7599, 3939, 7, 63, 10602, 41166,
1634, 11, 12718, 1115, 281, 8, 5, 854, 3964, 21785,
14113, 8, 63, 38905, 8, 21950, 19947, 11, 5, 1926,
6, 1625, 10902, 41, 5784, 4066, 3143, 9, 39456, 8,
856, 25729, 4, 993, 195, 5243, 66, 9, 262, 1360,
38950, 1848, 4707, 303, 11, 1625, 480, 5, 144, 11,
143, 247, 480, 64, 129, 28, 13590, 624, 63, 7562,
4, 85, 16, 184, 7, 39179, 3505, 9, 27772, 6,
28482, 4707, 9, 7723, 6, 112, 6, 6115, 17576, 9,
7723, 8, 973, 6, 151, 1380, 14868, 9, 3451, 4,
221, 2839, 8367, 102, 6, 10, 786, 12, 7699, 1651,
14, 1364, 7, 3720, 8360, 8, 5068, 709, 11, 1625,
6, 34, 3919, 411, 4707, 61, 24, 161, 7648, 2072,
5, 1272, 2713, 30, 5, 2],
[ 0, 36, 16256, 43, 480, 6871, 24, 1655, 22049, 50,
41571, 1896, 54, 24509, 13, 244, 5, 363, 5, 601,
12, 180, 12, 279, 1896, 21, 738, 1462, 116, 280,
115, 6723, 15, 61, 985, 5, 3940, 2046, 4, 1868,
22049, 18, 8, 1896, 18, 8826, 2327, 117, 28946, 273,
11, 2559, 461, 4961, 25, 7, 1060, 28604, 2236, 16,
1317, 11347, 148, 10, 7158, 486, 31, 14, 902, 973,
6, 1125, 6, 363, 11, 23058, 6, 1261, 35, 4028,
26, 24, 21, 69, 979, 4, 280, 34920, 480, 19,
5767, 3809, 1243, 21605, 13875, 24, 21, 69, 979, 6,
41571, 6, 54, 16670, 66, 6, 150, 15151, 2459, 22049,
26, 24, 21, 69, 979, 6, 1655, 6, 54, 21,
16600, 71, 145, 4487, 30, 5, 6066, 480, 21, 1353,
7, 273, 18, 461, 7069, 6, 8, 1353, 7, 5,
200, 12, 5743, 1900, 403, 25451, 11, 1353, 1261, 4,
22049, 34, 4407, 45, 2181, 8, 1695, 37, 738, 5,
7044, 11, 1403, 12, 21409, 4, 20, 7158, 486, 702,
2330, 11, 461, 15, 273, 6, 39, 3969, 2026, 6,
124, 62, 49, 19395, 14, 24, 21, 1896, 6, 8,
45, 49, 3653, 6, 54, 21, 5, 29593, 368, 4,
4500, 4945, 628, 273, 1390, 6, 15151, 2459, 22049, 26,
79, 21, 686, 1655, 21, 5, 65, 16600, 4, 2612,
116, 41039, 10105, 37, 18, 127, 979, 39732, 264, 7173,
41039, 1250, 9, 5, 1065, 48149, 77, 553, 549, 79,
56, 655, 137, 1317, 69, 1655, 22049, 7923, 24120, 50,
8930, 66, 13, 244, 4, 2]], device='cuda:0')
[[1]]$attention_mask
tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]], device='cuda:0')
[[1]]$decoder_input_ids
tensor([[ 0, 1625, 4452, 7, 62, 7, 158, 135, 9, 70,
684, 4707, 15, 3875, 479, 50118, 243, 16, 184, 7,
39179, 3505, 9, 27772, 6, 28482, 5103, 4707, 8, 973,
6, 151, 3505, 9, 3451, 479, 50118, 33837, 709, 8,
2147, 464, 16, 9405, 10, 380, 8793, 15, 63, 28809,
479, 50118, 133, 3274, 9587, 16, 223, 1856, 11, 14117,
9, 145, 5, 247, 18, 632, 7648, 479, 2, 1,
1, 1, 1, 1, 1, 1],
[ 0, 5178, 35, 83, 1443, 2470, 161, 97, 1283, 6,
45, 5, 7158, 486, 6, 40, 3094, 5, 403, 479,
50118, 5341, 35, 83, 2470, 13, 1896, 18, 284, 161,
37, 4265, 5, 3940, 40, 465, 22049, 2181, 479, 50118,
534, 15356, 2459, 22049, 161, 79, 2215, 5, 28604, 2236,
16, 14, 9, 69, 979, 479, 50118, 30541, 6, 41571,
1896, 18, 18631, 1843, 26, 14, 24, 69, 979, 18,
2236, 15, 5, 7158, 486, 479]], device='cuda:0')
[[1]]$labels
tensor([[ 1625, 4452, 7, 62, 7, 158, 135, 9, 70, 684,
4707, 15, 3875, 479, 50118, 243, 16, 184, 7, 39179,
3505, 9, 27772, 6, 28482, 5103, 4707, 8, 973, 6,
151, 3505, 9, 3451, 479, 50118, 33837, 709, 8, 2147,
464, 16, 9405, 10, 380, 8793, 15, 63, 28809, 479,
50118, 133, 3274, 9587, 16, 223, 1856, 11, 14117, 9,
145, 5, 247, 18, 632, 7648, 479, 2, -100, -100,
-100, -100, -100, -100, -100, -100],
[ 5178, 35, 83, 1443, 2470, 161, 97, 1283, 6, 45,
5, 7158, 486, 6, 40, 3094, 5, 403, 479, 50118,
5341, 35, 83, 2470, 13, 1896, 18, 284, 161, 37,
4265, 5, 3940, 40, 465, 22049, 2181, 479, 50118, 534,
15356, 2459, 22049, 161, 79, 2215, 5, 28604, 2236, 16,
14, 9, 69, 979, 479, 50118, 30541, 6, 41571, 1896,
18, 18631, 1843, 26, 14, 24, 69, 979, 18, 2236,
15, 5, 7158, 486, 479, 2]], device='cuda:0')
[[2]]
tensor([[ 1625, 4452, 7, 62, 7, 158, 135, 9, 70, 684,
4707, 15, 3875, 479, 50118, 243, 16, 184, 7, 39179,
3505, 9, 27772, 6, 28482, 5103, 4707, 8, 973, 6,
151, 3505, 9, 3451, 479, 50118, 33837, 709, 8, 2147,
464, 16, 9405, 10, 380, 8793, 15, 63, 28809, 479,
50118, 133, 3274, 9587, 16, 223, 1856, 11, 14117, 9,
145, 5, 247, 18, 632, 7648, 479, 2, -100, -100,
-100, -100, -100, -100, -100, -100],
[ 5178, 35, 83, 1443, 2470, 161, 97, 1283, 6, 45,
5, 7158, 486, 6, 40, 3094, 5, 403, 479, 50118,
5341, 35, 83, 2470, 13, 1896, 18, 284, 161, 37,
4265, 5, 3940, 40, 465, 22049, 2181, 479, 50118, 534,
15356, 2459, 22049, 161, 79, 2215, 5, 28604, 2236, 16,
14, 9, 69, 979, 479, 50118, 30541, 6, 41571, 1896,
18, 18631, 1843, 26, 14, 24, 69, 979, 18, 2236,
15, 5, 7158, 486, 479, 2]], device='cuda:0')
```
## Train
Define a ```Learner``` object and train the model:
```{r}
text_gen_kwargs = hf_config$task_specific_params['summarization'][[1]]
text_gen_kwargs['max_length'] = 130L; text_gen_kwargs['min_length'] = 30L
text_gen_kwargs
model = HF_BaseModelWrapper(hf_model)
model_cb = HF_SummarizationModelCallback(text_gen_kwargs=text_gen_kwargs)
learn = Learner(dls,
model,
opt_func=partial(Adam),
loss_func=CrossEntropyLossFlat(), #HF_PreCalculatedLoss()
cbs=model_cb,
splitter=partial(summarization_splitter, arch=hf_arch)) #.to_native_fp16() #.to_fp16()
learn$create_opt()
learn$freeze()
learn %>% fit_one_cycle(1, lr_max=4e-5)
```
## Conclusion
Predict a new text:
```{r}
test_article = c("About 10 men armed with pistols and small machine guns raided a casino in Switzerland
and made off into France with several hundred thousand Swiss francs in the early hours
of Sunday morning, police said. The men, dressed in black clothes and black ski masks,
split into two groups during the raid on the Grand Casino Basel, Chief Inspector Peter
Gill told CNN. One group tried to break into the casino's vault on the lower level
but could not get in, but they did rob the cashier of the money that was not secured,
he said. The second group of armed robbers entered the upper level where the roulette
and blackjack tables are located and robbed the cashier there, he said. As the thieves
were leaving the casino, a woman driving by and unaware of what was occurring unknowingly
blocked the armed robbers' vehicles. A gunman pulled the woman from her vehicle, beat
her, and took off for the French border. The other gunmen followed into France, which
is only about 100 meters (yards) from the casino, Gill said. There were about 600 people
in the casino at the time of the robbery. There were no serious injuries, although one
guest on the Casino floor was kicked in the head by one of the robbers when he moved,
the police officer said. Swiss authorities are working closely with French authorities,
Gill said. The robbers spoke French and drove vehicles with French lRicense plates.
CNN's Andreena Narayan contributed to this report.")
```
```{r}
outputs = learn$blurr_summarize(test_article, num_return_sequences=3L)
cat(outputs)
```
```
Robbers made off with several hundred thousand Swiss francs in the early hours of Sunday morning, police say .
The men, dressed in black clothes and black ski masks, split into two groups during the raid on the Grand Casino Basel .
There were no serious injuries, although one guest was kicked in the head by one of the robbers when he moved, police officer says .
A woman driving by unknowingly blocked the robbers' vehicles and was beaten to death .
Swiss authorities are working closely with French authorities, police chief says .
Robbers made off with several hundred thousand Swiss francs in the early hours of Sunday morning, police say .
The men, dressed in black clothes and black ski masks, split into two groups during the raid on the Grand Casino Basel .
There were no serious injuries, although one guest was kicked in the head by one of the robbers when he moved, police officer says .
A woman driving by unknowingly blocked the robbers' vehicles and was beaten to death .
Swiss authorities are working closely with French authorities, an officer says.
Robbers made off with several hundred thousand Swiss francs in the early hours of Sunday morning, police say .
The men, dressed in black clothes and black ski masks, split into two groups during the raid on the Grand Casino Basel .
There were no serious injuries, although one guest was kicked in the head by one of the robbers when he moved, police officer says .
A woman driving by unknowingly blocked the robbers' vehicles and was beaten to death .
```