Commit
·
c72e58a
1
Parent(s):
074d9b5
Update README.md
Browse files
README.md
CHANGED
|
@@ -52,12 +52,17 @@ training_args = TrainingArguments(
|
|
| 52 |
### Eval results
|
| 53 |
The model was evaluated using the test sets for MultiNLI and ANLI and the dev set for Fever-NLI
|
| 54 |
dataset | accuracy
|
| 55 |
-
|
| 56 |
-
|
|
|
|
| 57 |
fever-nli | 0.777
|
| 58 |
anli-all | 0.579
|
| 59 |
anli-r3 | 0.495
|
| 60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
## Limitations and bias
|
| 63 |
Please consult the original DeBERTa paper and literature on different NLI datasets for potential biases.
|
|
|
|
| 52 |
### Eval results
|
| 53 |
The model was evaluated using the test sets for MultiNLI and ANLI and the dev set for Fever-NLI
|
| 54 |
dataset | accuracy
|
| 55 |
+
-------|------
|
| 56 |
+
mnli-m | 0.903
|
| 57 |
+
mnli-mm | 0.903
|
| 58 |
fever-nli | 0.777
|
| 59 |
anli-all | 0.579
|
| 60 |
anli-r3 | 0.495
|
| 61 |
|
| 62 |
+
metric | | precision | recall | accuracy (not balanced)
|
| 63 |
+
-------|---------|----------|---------|----------
|
| 64 |
+
accuracy | 0.773 | 0.772 | 0.771 | 0.771
|
| 65 |
+
|
| 66 |
|
| 67 |
## Limitations and bias
|
| 68 |
Please consult the original DeBERTa paper and literature on different NLI datasets for potential biases.
|