luisespinosa
commited on
Commit
ยท
8b48c2c
1
Parent(s):
9ad0c1d
update readme
Browse files
README.md
CHANGED
@@ -5,4 +5,43 @@ This is a roBERTa-base model trained on ~58M tweets, described and evaluated in
|
|
5 |
|
6 |
## Ejemplo MLM
|
7 |
|
8 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
## Ejemplo MLM
|
7 |
|
8 |
+
```python
|
9 |
+
from transformers import pipeline, AutoTokenizer
|
10 |
+
import numpy as np
|
11 |
+
|
12 |
+
MODEL = "cardiffnlp/roberta-base-rt"
|
13 |
+
fill_mask = pipeline("fill-mask", model=MODEL, tokenizer=MODEL)
|
14 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL)
|
15 |
+
|
16 |
+
def print_candidates():
|
17 |
+
for i in range(5):
|
18 |
+
token = tokenizer.decode(candidates[i]['token'])
|
19 |
+
score = np.round(candidates[i]['score'], 4)
|
20 |
+
print(f"{i+1}) {token} {score}")
|
21 |
+
|
22 |
+
texts = [
|
23 |
+
"I am so <mask> ๐",
|
24 |
+
"I am so <mask> ๐ข"
|
25 |
+
]
|
26 |
+
for text in texts:
|
27 |
+
print(f"{'-'*30}\n{text}")
|
28 |
+
candidates = fill_mask(text)
|
29 |
+
print_candidates()
|
30 |
+
```
|
31 |
+
|
32 |
+
```
|
33 |
+
------------------------------
|
34 |
+
I am so <mask> ๐
|
35 |
+
1) happy 0.402
|
36 |
+
2) excited 0.1441
|
37 |
+
3) proud 0.143
|
38 |
+
4) grateful 0.0669
|
39 |
+
5) blessed 0.0334
|
40 |
+
------------------------------
|
41 |
+
I am so <mask> ๐ข
|
42 |
+
1) sad 0.2641
|
43 |
+
2) sorry 0.1605
|
44 |
+
3) tired 0.138
|
45 |
+
4) sick 0.0278
|
46 |
+
5) hungry 0.0232
|
47 |
+
```
|