Update README.md
Browse files
README.md
CHANGED
|
@@ -24,6 +24,14 @@ It achieved 35.66%+-0.85%/51.68+-0.63% on MMLU-Da in [ScandEval](https://github.
|
|
| 24 |
|
| 25 |
Further evaluations will be tested.
|
| 26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
# Datasets
|
| 28 |
This model is trained on Danish instruction datasets [danish-OpenHermes](Mabeck/danish-OpenHermes) and [skoleGPT](https://huggingface.co/datasets/kobprof/skolegpt-instruct), which have not been safeguarded or alligned.
|
| 29 |
|
|
|
|
| 24 |
|
| 25 |
Further evaluations will be tested.
|
| 26 |
|
| 27 |
+
# Benchmarks
|
| 28 |
+
|
| 29 |
+
The following benchmarks have been tested using [ScandEval](https://github.com/ScandEval/ScandEval). Rankings don't include merged models and GPT4 and GPT3.5 ranks 1 and 2:
|
| 30 |
+
|
| 31 |
+
- **MMLU-da**: 35.66%+-0.85%/51.68+-0.63%, ranks 3rd
|
| 32 |
+
- **DANSK**: 50.77%+-2.29%/34.05%+-1.78%, ranks 3rd=
|
| 33 |
+
- **Hellaswag-da**: 29.18%+-0.99%/46.64%+-0.76%, ranks 4th
|
| 34 |
+
|
| 35 |
# Datasets
|
| 36 |
This model is trained on Danish instruction datasets [danish-OpenHermes](Mabeck/danish-OpenHermes) and [skoleGPT](https://huggingface.co/datasets/kobprof/skolegpt-instruct), which have not been safeguarded or alligned.
|
| 37 |
|