Datasets:
Update README.md
#5
by
Stopwolf
- opened
README.md
CHANGED
|
@@ -35,7 +35,7 @@ We then compare likelihoods of each letter (`A, B, C, D, E`) and calculate the f
|
|
| 35 |
|
| 36 |
GPT-like models were evaluated by taking top 20 probabilities of the first output token, which were further filtered for letters `A` to `E`. Letter with the highest probability was taken as a final answer.
|
| 37 |
|
| 38 |
-
Exact code for the task can be found
|
| 39 |
|
| 40 |
You should run the evaluation with the following command (do not forget to add --use_chat_template):
|
| 41 |
```
|
|
|
|
| 35 |
|
| 36 |
GPT-like models were evaluated by taking top 20 probabilities of the first output token, which were further filtered for letters `A` to `E`. Letter with the highest probability was taken as a final answer.
|
| 37 |
|
| 38 |
+
Exact code for the task can be found [here](https://github.com/Stopwolf/lighteval/blob/oz-eval/community_tasks/oz_evals.py).
|
| 39 |
|
| 40 |
You should run the evaluation with the following command (do not forget to add --use_chat_template):
|
| 41 |
```
|