Remove agreement-level subset section from README
Browse files
README.md
CHANGED
|
@@ -57,15 +57,6 @@ Training pipeline:
|
|
| 57 |
3. Apply task-adaptive pretraining on financial task text
|
| 58 |
4. Fine-tune for 3-class sentiment classification
|
| 59 |
|
| 60 |
-
For supervised fine-tuning, we evaluated multiple agreement-level subsets derived from Financial PhraseBank-style annotations:
|
| 61 |
-
|
| 62 |
-
- `100%` agreement
|
| 63 |
-
- `75%+` agreement
|
| 64 |
-
- `66%+` agreement
|
| 65 |
-
- `50%+` agreement
|
| 66 |
-
|
| 67 |
-
The final model released here is the `66%+` version.
|
| 68 |
-
|
| 69 |
## Evaluation Summary
|
| 70 |
|
| 71 |
Validation results for the selected `75%+` configuration:
|
|
|
|
| 57 |
3. Apply task-adaptive pretraining on financial task text
|
| 58 |
4. Fine-tune for 3-class sentiment classification
|
| 59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
## Evaluation Summary
|
| 61 |
|
| 62 |
Validation results for the selected `75%+` configuration:
|