Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language: it
|
| 3 |
+
license: apache-2.0
|
| 4 |
+
widget:
|
| 5 |
+
- text: "Il [MASK] ha chiesto revocarsi l'obbligo di pagamento"
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
# ITALIAN-LEGAL-BERT-SC
|
| 9 |
+
It is the [ITALIAN-LEGAL-BERT](https://huggingface.co/dlicari/Italian-Legal-BERT) variant pre-trained from scratch on Italian legal documents (ITA-LEGAL-BERT-SC) based on the CamemBERT architecture
|
| 10 |
+
|
| 11 |
+
## Training procedure
|
| 12 |
+
It was trained from scratch using a larger training dataset, 6.6GB of civil and criminal cases.
|
| 13 |
+
We used [CamemBERT](https://huggingface.co/docs/transformers/main/en/model_doc/camembert) architecture with a language modeling head on top, AdamW Optimizer, initial learning rate 2e-5 (with linear learning rate decay), sequence length 512, batch size 18, 1 million training steps,
|
| 14 |
+
device 8*NVIDIA A100 40GB using distributed data parallel (each step performs 8 batches). It uses SentencePiece tokenization trained from scratch on a subset of training set (5 milions sentences)
|
| 15 |
+
and vocabulary size of 32000
|
| 16 |
+
|
| 17 |
+
|