Fix sorting heuristic
#3
by
Markus28
- opened
- modeling_bert.py +3 -2
modeling_bert.py
CHANGED
|
@@ -1211,8 +1211,9 @@ class JinaBertModel(JinaBertPreTrainedModel):
|
|
| 1211 |
self.to(device)
|
| 1212 |
|
| 1213 |
# TODO: Maybe use better length heuristic?
|
| 1214 |
-
|
| 1215 |
-
|
|
|
|
| 1216 |
|
| 1217 |
padding = tokenizer_kwargs.pop('padding', True)
|
| 1218 |
|
|
|
|
| 1211 |
self.to(device)
|
| 1212 |
|
| 1213 |
# TODO: Maybe use better length heuristic?
|
| 1214 |
+
permutation = np.argsort([-len(i) for i in sentences])
|
| 1215 |
+
inverse_permutation = np.argsort(permutation)
|
| 1216 |
+
sentences = [sentences[idx] for idx in permutation]
|
| 1217 |
|
| 1218 |
padding = tokenizer_kwargs.pop('padding', True)
|
| 1219 |
|