sberbank-ai
commited on
Commit
•
499e3bf
1
Parent(s):
5198f5c
Update README.md
Browse files
README.md
CHANGED
@@ -1,4 +1,21 @@
|
|
1 |
---
|
2 |
language:
|
3 |
- ru
|
4 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
language:
|
3 |
- ru
|
4 |
+
---
|
5 |
+
|
6 |
+
# FRED-T5 1.7B (Full-scale Russian Enhanced Denoisers T5)
|
7 |
+
|
8 |
+
Architecture based on T5.
|
9 |
+
|
10 |
+
It has 24 layers and 1536 hidden size.
|
11 |
+
|
12 |
+
Model was trained on a mixture of 7 denoisers like UL2 with several differences .
|
13 |
+
|
14 |
+
It trained on Russian language corpus (300GB). The dataset is the same as for ruT5 models.
|
15 |
+
|
16 |
+
Bbpe tokenizer. First half of the time model was trained on the small part of all datasets (1%).
|
17 |
+
|
18 |
+
|
19 |
+
We continue to experiment...
|
20 |
+
|
21 |
+
We'll tell you more and release checkpoint to the public soon.
|