---
license: apache-2.0
library_name: transformers
---
# ScienceGPT
<!-- markdownlint-disable first-line-h1 -->
<!-- markdownlint-disable html -->
<!-- markdownlint-disable no-duplicate-header -->

<div align="center">
  <img src="figures/fig1.png" width="60%" alt="ScienceGPT" />
</div>
<hr>

<div align="center" style="line-height: 1;">
  <a href="LICENSE" style="margin: 2px;">
    <img alt="License" src="figures/fig2.png" style="display: inline-block; vertical-align: middle;"/>
  </a>
</div>

## 1. Introduction

ScienceGPT is a specialized language model fine-tuned for scientific reasoning and knowledge. The model has been trained on extensive scientific literature and datasets, demonstrating exceptional capabilities in physics, chemistry, biology, mathematics, and earth sciences. It excels at solving complex scientific problems and explaining scientific concepts.

<p align="center">
  <img width="80%" src="figures/fig3.png">
</p>

The model shows remarkable improvements in scientific domain understanding compared to general-purpose models. For instance, in standardized science examinations, ScienceGPT achieves 85% accuracy compared to 65% for baseline models.

## 2. Evaluation Results

### Comprehensive Benchmark Results

<div align="center">

| | Benchmark | BaseModel | Model-v1 | Model-v2 | ScienceGPT |
|---|---|---|---|---|---|
| **Core Sciences** | Physics | 0.620 | 0.645 | 0.660 | 0.593 |
| | Chemistry | 0.580 | 0.595 | 0.610 | 0.628 |
| | Biology | 0.640 | 0.670 | 0.685 | 0.735 |
| | Mathematics | 0.710 | 0.735 | 0.750 | 0.727 |
| | Earth Science | 0.590 | 0.615 | 0.630 | 0.661 |

</div>

### Overall Performance Summary
ScienceGPT demonstrates strong performance across all scientific benchmark categories, with particularly notable results in mathematics and biology.

## 3. Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("username/ScienceGPT-TestRepo")
tokenizer = AutoTokenizer.from_pretrained("username/ScienceGPT-TestRepo")
```

## 4. License
This model is licensed under the [Apache 2.0 License](LICENSE).

## 5. Contact
For questions, please contact science@example.com.