Update README.md
Browse files
README.md
CHANGED
|
@@ -36,6 +36,7 @@ tokeniser.batch_decode(outputs, skip_special_tokens=True, clean_up_tokenisation_
|
|
| 36 |
|
| 37 |
The models' predictions should not be taken as ground-truth *titrages* and should always be indicated as being automatically generated. They were designed not to be used as such, but to improve search coverage for improved similarity prediction between different cases (the predicted *titrages* being used to predict the similarity).
|
| 38 |
|
|
|
|
| 39 |
|
| 40 |
## Training data
|
| 41 |
|
|
|
|
| 36 |
|
| 37 |
The models' predictions should not be taken as ground-truth *titrages* and should always be indicated as being automatically generated. They were designed not to be used as such, but to improve search coverage for improved similarity prediction between different cases (the predicted *titrages* being used to predict the similarity).
|
| 38 |
|
| 39 |
+
The model is not constrained to predict *titres* that have previously been seen, so this should be taken into account in the deployment of this model as a *titrage* tool in order to avoid the multiplication of different *titres*.
|
| 40 |
|
| 41 |
## Training data
|
| 42 |
|