Commit
•
be66bc0
1
Parent(s):
6607ffa
Added readme file
Browse files
README.md
ADDED
|
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
datasets:
|
| 3 |
+
- oscar
|
| 4 |
+
- srwac
|
| 5 |
+
- leipzig
|
| 6 |
+
- cc100
|
| 7 |
+
- hrwac
|
| 8 |
+
|
| 9 |
+
language:
|
| 10 |
+
- hr
|
| 11 |
+
- sr
|
| 12 |
+
|
| 13 |
+
tags:
|
| 14 |
+
- masked-lm
|
| 15 |
+
|
| 16 |
+
widget:
|
| 17 |
+
- text: "Ovo je početak <mask>."
|
| 18 |
+
|
| 19 |
+
license: apache-2.0
|
| 20 |
+
|
| 21 |
+
---
|
| 22 |
+
# Transformer language model for Croatian and Serbian
|
| 23 |
+
Trained on 43GB datasets that contain Croatian and Serbian language for one epochs (9.6 mil. steps, 3 epochs).
|
| 24 |
+
Leipzig Corpus, OSCAR, srWac, hrWac, cc100-hr and cc100-sr datasets
|
| 25 |
+
|
| 26 |
+
Validation number of exampels run for perplexity:1620487 sentences
|
| 27 |
+
Perplexity:6.02
|
| 28 |
+
Start loss: 8.6
|
| 29 |
+
Final loss: 2.0
|
| 30 |
+
Thoughts: Model could be trained more, the training did not stagnate.
|
| 31 |
+
|
| 32 |
+
| Model | #params | Arch. | Training data |
|
| 33 |
+
|--------------------------------|--------------------------------|-------|-----------------------------------|
|
| 34 |
+
| `Andrija/SRoBERTa-X` | 80M | Fifth | Leipzig Corpus, OSCAR, srWac, hrWac, cc100-hr and cc100-sr (43 GB of text) |
|