Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -6,14 +6,17 @@ task_categories:
|
|
6 |
- question-answering
|
7 |
---
|
8 |
|
9 |
-
|
|
|
|
|
|
|
10 |
OZ Eval (_sr._ Opšte Znanje Evaluacija) dataset was created for the purposes of evaluating General Knowledge of LLM models in Serbian language.
|
11 |
Data consists of 1k+ high-quality questions and answers which were used as part of entry exams at the Faculty of Philosophy and Faculty of Organizational Sciences, University of Belgrade.
|
12 |
The exams test the General Knowledge of students and were used in the enrollment periods from 2003 to 2024.
|
13 |
|
14 |
|
15 |
## Evaluation process
|
16 |
-
Models are evaluated by the following principle using HuggingFace's library lighteval
|
17 |
```
|
18 |
Pitanje: {question}
|
19 |
|
@@ -26,7 +29,9 @@ E. {option_e}
|
|
26 |
|
27 |
Krajnji odgovor:
|
28 |
```
|
29 |
-
We then compare likelihoods of each letter (`A, B, C, D, E`) and calculate the final accuracy. All evaluations are ran in a 0-shot manner using a chat template
|
|
|
|
|
30 |
|
31 |
Exact code for the task will be posted [here]().
|
32 |
|
@@ -34,9 +39,13 @@ Exact code for the task will be posted [here]().
|
|
34 |
## Evaluation results
|
35 |
| Model |Accuracy| |Stderr|
|
36 |
|-------|-------:|--|-----:|
|
|
|
|
|
|
|
37 |
|[Tito-7B-slerp](https://huggingface.co/Stopwolf/Tito-7B-slerp)|0.7099|±|0.0101|
|
38 |
[Qwen2-7B-instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct)|0.673|±|0.0105|
|
39 |
|[Yugo60-GPT](https://huggingface.co/datatab/Yugo60-GPT)|0.6411|±|0.0107|
|
|
|
40 |
|[Llama3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct)|0.5274|±|0.0111|
|
41 |
|[Hermes-2-Pro-Mistral-7B](https://huggingface.co/NousResearch/Hermes-2-Pro-Mistral-7B)|0.5145|±|0.0112|
|
42 |
|[Perucac-7B-slerp](https://huggingface.co/Stopwolf/Perucac-7B-slerp)|0.4247|±|0.011|
|
|
|
6 |
- question-answering
|
7 |
---
|
8 |
|
9 |
+
# OZ Eval
|
10 |
+
![image/png](https://cdn-uploads.huggingface.co/production/uploads/62d01cf580c9d4ceb20254bb/Ft9w9DNRN_1bzuQC9XNSG.png)
|
11 |
+
|
12 |
+
## Dataset Description
|
13 |
OZ Eval (_sr._ Opšte Znanje Evaluacija) dataset was created for the purposes of evaluating General Knowledge of LLM models in Serbian language.
|
14 |
Data consists of 1k+ high-quality questions and answers which were used as part of entry exams at the Faculty of Philosophy and Faculty of Organizational Sciences, University of Belgrade.
|
15 |
The exams test the General Knowledge of students and were used in the enrollment periods from 2003 to 2024.
|
16 |
|
17 |
|
18 |
## Evaluation process
|
19 |
+
Models are evaluated by the following principle using HuggingFace's library `lighteval`. We supply the model with the following template:
|
20 |
```
|
21 |
Pitanje: {question}
|
22 |
|
|
|
29 |
|
30 |
Krajnji odgovor:
|
31 |
```
|
32 |
+
We then compare likelihoods of each letter (`A, B, C, D, E`) and calculate the final accuracy. All evaluations are ran in a 0-shot manner using a **chat template**.
|
33 |
+
|
34 |
+
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.
|
35 |
|
36 |
Exact code for the task will be posted [here]().
|
37 |
|
|
|
39 |
## Evaluation results
|
40 |
| Model |Accuracy| |Stderr|
|
41 |
|-------|-------:|--|-----:|
|
42 |
+
|GPT-4-0125-preview|0.9199|±|0.002|
|
43 |
+
|GPT-4o-2024-05-13|0.9196|±|0.0017|
|
44 |
+
|GPT-3.5-turbo-0125|0.8245|±|0.0016|
|
45 |
|[Tito-7B-slerp](https://huggingface.co/Stopwolf/Tito-7B-slerp)|0.7099|±|0.0101|
|
46 |
[Qwen2-7B-instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct)|0.673|±|0.0105|
|
47 |
|[Yugo60-GPT](https://huggingface.co/datatab/Yugo60-GPT)|0.6411|±|0.0107|
|
48 |
+
|[Hermes-2-Theta-Llama-3-8B](https://huggingface.co/NousResearch/Hermes-2-Theta-Llama-3-8B)|0.5852|±|0.011|
|
49 |
|[Llama3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct)|0.5274|±|0.0111|
|
50 |
|[Hermes-2-Pro-Mistral-7B](https://huggingface.co/NousResearch/Hermes-2-Pro-Mistral-7B)|0.5145|±|0.0112|
|
51 |
|[Perucac-7B-slerp](https://huggingface.co/Stopwolf/Perucac-7B-slerp)|0.4247|±|0.011|
|