Commit
•
9588374
1
Parent(s):
74d70b9
Update README.md
Browse files
README.md
CHANGED
|
@@ -13,12 +13,11 @@ widget:
|
|
| 13 |
---
|
| 14 |
This is an updated version of [cointegrated/rubert-tiny](https://huggingface.co/cointegrated/rubert-tiny): a small Russian BERT-based encoder with high-quality sentence embeddings.
|
| 15 |
|
| 16 |
-
**DISCLAIMER: the model is going to be updated, and the current version is unstable.**
|
| 17 |
-
|
| 18 |
The differences from the previous version include:
|
| 19 |
- a larger vocabulary: 83828 tokens instead of 29564;
|
| 20 |
- larger supported sequences: 2048 instead of 512;
|
| 21 |
- sentence embeddings approximate LaBSE closer than before;
|
|
|
|
| 22 |
- the model is focused only on Russian.
|
| 23 |
|
| 24 |
The model should be used as is to produce sentence embeddings (e.g. for KNN classification of short texts) or fine-tuned for a downstream task.
|
| 13 |
---
|
| 14 |
This is an updated version of [cointegrated/rubert-tiny](https://huggingface.co/cointegrated/rubert-tiny): a small Russian BERT-based encoder with high-quality sentence embeddings.
|
| 15 |
|
|
|
|
|
|
|
| 16 |
The differences from the previous version include:
|
| 17 |
- a larger vocabulary: 83828 tokens instead of 29564;
|
| 18 |
- larger supported sequences: 2048 instead of 512;
|
| 19 |
- sentence embeddings approximate LaBSE closer than before;
|
| 20 |
+
- meaningful segment embeddings (tuned on the NLI task)
|
| 21 |
- the model is focused only on Russian.
|
| 22 |
|
| 23 |
The model should be used as is to produce sentence embeddings (e.g. for KNN classification of short texts) or fine-tuned for a downstream task.
|