add a bit more info to model card
Browse files
README.md
CHANGED
@@ -5,10 +5,10 @@ license: mit
|
|
5 |
thumbnail: https://en.wikipedia.org/wiki/Bart_Simpson#/media/File:Bart_Simpson_200px.png
|
6 |
---
|
7 |
# BART for Gigaword
|
8 |
-
- This model was created by fine-tuning the facebook/bart-large-cnn weights (also on HuggingFace) for the Gigaword dataset. The model was fine-tuned on the Gigaword training set for 3 epochs, and the model with the highest ROUGE-1 score on the training set batches was kept.
|
9 |
- The BART Tokenizer for CNN-Dailymail was used in the fine-tuning process and that is the tokenizer that will be loaded automatically when doing:
|
10 |
```
|
11 |
from transformers import AutoTokenizer
|
12 |
tokenizer = AutoTokenizer.from_pretrained("a1noack/bart-large-gigaword")
|
13 |
```
|
14 |
-
- This model achieves ROUGE-1 / ROUGE-2 / ROUGE-L of 37.28 / 18.58 / 34.53 on the Gigaword test set; this is pretty good when compared to
|
|
|
5 |
thumbnail: https://en.wikipedia.org/wiki/Bart_Simpson#/media/File:Bart_Simpson_200px.png
|
6 |
---
|
7 |
# BART for Gigaword
|
8 |
+
- This model was created by fine-tuning the `facebook/bart-large-cnn` weights (also on HuggingFace) for the Gigaword dataset. The model was fine-tuned on the Gigaword training set for 3 epochs, and the model with the highest ROUGE-1 score on the training set batches was kept.
|
9 |
- The BART Tokenizer for CNN-Dailymail was used in the fine-tuning process and that is the tokenizer that will be loaded automatically when doing:
|
10 |
```
|
11 |
from transformers import AutoTokenizer
|
12 |
tokenizer = AutoTokenizer.from_pretrained("a1noack/bart-large-gigaword")
|
13 |
```
|
14 |
+
- This model achieves ROUGE-1 / ROUGE-2 / ROUGE-L of 37.28 / 18.58 / 34.53 on the Gigaword test set; this is pretty good when compared to PEGASUS results of 39.12 / 19.86 / 36.24.
|