Update README.md
Browse files
README.md
CHANGED
|
@@ -222,6 +222,7 @@ Moreover, it outperforms all other Russian-language models present in this bench
|
|
| 222 |
|
| 223 |
It is particularly worth noting the significant progress on the CheGeKa dataset. Although the data we used for further training did not include examples from this or similar games, we observed more than a 4x increase in the F1 score (0.071 -> 0.305). Additionally, the EM (Exact Match) metric improved from 0 to 0.231. On this dataset, our model significantly outperforms other Russian-language models as well.
|
| 224 |
|
|
|
|
| 225 |
|
| 226 |
## Example Usage
|
| 227 |
|
|
|
|
| 222 |
|
| 223 |
It is particularly worth noting the significant progress on the CheGeKa dataset. Although the data we used for further training did not include examples from this or similar games, we observed more than a 4x increase in the F1 score (0.071 -> 0.305). Additionally, the EM (Exact Match) metric improved from 0 to 0.231. On this dataset, our model significantly outperforms other Russian-language models as well.
|
| 224 |
|
| 225 |
+
We have also significantly reduced the number of switches from Russian to English. In our internal benchmark, the original model switched 32% of the time, while the current model does so only 0.7% of the time.
|
| 226 |
|
| 227 |
## Example Usage
|
| 228 |
|