Updated LLM eval leaderboard in Read.me
Browse files
README.md
CHANGED
@@ -52,6 +52,7 @@ dataset_info:
|
|
52 |
num_bytes: 34663548
|
53 |
num_examples: 614
|
54 |
download_size: 69119656
|
|
|
55 |
dataset_size: 69327096.0
|
56 |
tags:
|
57 |
- mathematics
|
@@ -74,9 +75,8 @@ language:
|
|
74 |
- en
|
75 |
- ms
|
76 |
---
|
77 |
-
**STEM_Dataset_eng_ms**
|
78 |
|
79 |
-
**A Bilingual Dataset for Evaluating Reasoning Skills in STEM Subjects**
|
80 |
|
81 |
This dataset provides a comprehensive evaluation set for tasks assessing reasoning skills in Science, Technology, Engineering, and Mathematics (STEM) subjects. It features questions in both English and Malay, catering to a diverse audience.
|
82 |
|
@@ -103,6 +103,8 @@ The dataset is comprised of two configurations: `data_en` (English) and `data_ms
|
|
103 |
* **Options:** Possible answer choices for the question, with keys (e.g., "A", "B", "C", "D") and corresponding text.
|
104 |
* **Answers:** Correct answer to the question, represented by the key of the correct option (e.g., "C").
|
105 |
|
|
|
|
|
106 |
## Data Instance Example
|
107 |
|
108 |
```json
|
@@ -161,11 +163,15 @@ The dataset is derived from a combination of resources, including:
|
|
161 |
* **Release Date:** December 27, 2024
|
162 |
* **Contact:** We welcome any feedback or corrections to improve the dataset quality.
|
163 |
|
164 |
-
|
|
|
|
|
165 |
|
166 |
This dataset is licensed under the [Creative Commons Attribution 4.0 International License (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/).
|
167 |
|
168 |
-
|
|
|
|
|
169 |
|
170 |
You can access the dataset on Hugging Face using the following commands:
|
171 |
|
@@ -180,6 +186,52 @@ dataset = load_dataset("Supa-AI/STEM-en-ms", name="data_en")
|
|
180 |
# For Malay data
|
181 |
dataset = load_dataset("Supa-AI/STEM-en-ms", name="data_ms")
|
182 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
|
184 |
**Contributors**
|
185 |
- [Gele](https://huggingface.co/Geleliong)
|
|
|
52 |
num_bytes: 34663548
|
53 |
num_examples: 614
|
54 |
download_size: 69119656
|
55 |
+
|
56 |
dataset_size: 69327096.0
|
57 |
tags:
|
58 |
- mathematics
|
|
|
75 |
- en
|
76 |
- ms
|
77 |
---
|
|
|
78 |
|
79 |
+
# **A Bilingual Dataset for Evaluating Reasoning Skills in STEM Subjects**
|
80 |
|
81 |
This dataset provides a comprehensive evaluation set for tasks assessing reasoning skills in Science, Technology, Engineering, and Mathematics (STEM) subjects. It features questions in both English and Malay, catering to a diverse audience.
|
82 |
|
|
|
103 |
* **Options:** Possible answer choices for the question, with keys (e.g., "A", "B", "C", "D") and corresponding text.
|
104 |
* **Answers:** Correct answer to the question, represented by the key of the correct option (e.g., "C").
|
105 |
|
106 |
+
---
|
107 |
+
|
108 |
## Data Instance Example
|
109 |
|
110 |
```json
|
|
|
163 |
* **Release Date:** December 27, 2024
|
164 |
* **Contact:** We welcome any feedback or corrections to improve the dataset quality.
|
165 |
|
166 |
+
---
|
167 |
+
|
168 |
+
# License
|
169 |
|
170 |
This dataset is licensed under the [Creative Commons Attribution 4.0 International License (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/).
|
171 |
|
172 |
+
---
|
173 |
+
|
174 |
+
# Getting Started
|
175 |
|
176 |
You can access the dataset on Hugging Face using the following commands:
|
177 |
|
|
|
186 |
# For Malay data
|
187 |
dataset = load_dataset("Supa-AI/STEM-en-ms", name="data_ms")
|
188 |
```
|
189 |
+
---
|
190 |
+
|
191 |
+
# Bilingual STEM Dataset LLM Leaderboard
|
192 |
+
|
193 |
+
This document summarizes the evaluation results for various language models based on **5-shot** and **First Token Accuracy**. The evaluation was conducted across four configurations:
|
194 |
+
|
195 |
+
- **en_withfigures**: English prompts including figures.
|
196 |
+
- **en_withoutfigures**: English prompts excluding figures.
|
197 |
+
- **ms_withfigures**: Malay prompts including figures.
|
198 |
+
- **ms_withoutfigures**: Malay prompts excluding figures.
|
199 |
+
|
200 |
+
---
|
201 |
+
|
202 |
+
## Results Table
|
203 |
+
|
204 |
+
| **Model** | **en_withfigures** | **en_withoutfigures** | **ms_withfigures** | **ms_withoutfigures** |
|
205 |
+
|---------------------------------|--------------------|-----------------------|--------------------|-----------------------|
|
206 |
+
| **gemini-2.0-flash-exp** | __63.70%__ | **75.16%** | __63.36%__ | **75.47%** |
|
207 |
+
| **gemini-1.5-flash** | __49.66%__ | __67.39%__ | __50.00%__ | __64.28%__ |
|
208 |
+
| **Qwen/Qwen2-VL-72B-Instruct** | __58.22%__ | __69.25%__ | __57.53%__ | __63.66%__ |
|
209 |
+
| **gpt-4o** | __47.95%__ | __66.15%__ | __50.00%__ | __68.01%__ |
|
210 |
+
| **gpt-4o-mini** | __41.10%__ | __55.90%__ | __38.36%__ | __52.80%__ |
|
211 |
+
| **pixtral-large-2411** | __42.81%__ | __64.29%__ | __35.27%__ | __60.87%__ |
|
212 |
+
| **pixtral-12b-2409** | __24.66%__ | __48.45%__ | __24.66%__ | __39.13%__ |
|
213 |
+
| **DeepSeek-V3** | None | **79.19%** | None | **76.40%** |
|
214 |
+
| **Qwen2.5-72B-Instruct** | None | __74.53%__ | None | __72.98%__ |
|
215 |
+
| **Meta-Llama-3.3-70B-Instruct** | None | __67.08%__ | None | __58.07%__ |
|
216 |
+
| **Llama-3.2-90B-Vision-Instruct** | None | __65.22%__ | None | __58.07%__ |
|
217 |
+
| **sail/Sailor2-20B-Chat** | None | __66.46%__ | None | __61.68%__ |
|
218 |
+
| **mallam-small** | None | __61.49%__ | None | __55.28%__ |
|
219 |
+
| **mistral-large-latest** | None | __60.56%__ | None | __53.42%__ |
|
220 |
+
| **google/gemma-2-27b-it** | None | __58.07%__ | None | __57.76%__ |
|
221 |
+
| **SeaLLMs-v3-7B-Chat** | None | __50.93%__ | None | __45.96%__ |
|
222 |
+
|
223 |
+
---
|
224 |
+
|
225 |
+
## Notes
|
226 |
+
|
227 |
+
In the repository, there is an `eval.py` script that can be used to run the evaluation for any other LLM.
|
228 |
+
|
229 |
+
|
230 |
+
- The evaluation results are based on the specific dataset and methodology employed.
|
231 |
+
- The "First Token Accuracy" metric emphasizes the accuracy of predicting the initial token correctly.
|
232 |
+
- Further analysis might be needed to determine the models' suitability for specific tasks.
|
233 |
+
|
234 |
+
---
|
235 |
|
236 |
**Contributors**
|
237 |
- [Gele](https://huggingface.co/Geleliong)
|