DjMel commited on
Commit
86b1104
1 Parent(s): ace1bc7

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +12 -3
README.md CHANGED
@@ -6,14 +6,17 @@ task_categories:
6
  - question-answering
7
  ---
8
 
9
- ## OZ Eval
 
 
 
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. We supply the model with the following template:
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|