---
license: apache-2.0
base_model: Qwen/Qwen2.5-0.5B-Instruct
library_name: transformers
tags:
- science
- domain-adaptation
- fine-tuned
- evaluation:gpqa
pipeline_tag: text-generation
language:
- en
datasets:
- custom-science-corpus
model-index:
- name: Science Pharaoh Mini
  results:
  - task:
      type: text-generation
      name: Multiple Choice QA
    dataset:
      name: GPQA Diamond
      type: gpqa_diamond
    metrics:
    - name: Accuracy (Single-Token MC)
      type: accuracy
      value: 34.8
---

# Science Pharaoh Mini

**Science Pharaoh Mini** is a specialized 0.5-parameter language model fine-tuned on scientific reasoning and domain-specific knowledge. Built upon Alibaba Cloud's [Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct), this model demonstrates optimized performance on graduate-level science reasoning tasks relative to its lightweight footprint.

---

## Key Highlights & Performance

- **Model Size:** 0.5B parameters (Compact, edge-ready)
- **Target Domain:** General Science, Physics, Chemistry, Biology
- **Benchmark Performance:** Reached **34.8%** accuracy on the highly challenging GPQA Diamond benchmark using a single-token evaluation harness.

### Benchmark Evaluation

| Benchmark | Setting | Score | Total Questions | Accuracy |
| :--- | :--- | :--- | :--- | :--- |
| **GPQA Diamond** | Single-Token MC (0-Shot) | 69 / 198 | 198 | **34.8%** |

---

## Quickstart & Usage

### Setup & Directory Layout
The repository artifact package includes all necessary files organized as follows:
```text
.
└── final_model/
    ├── config.json
    ├── model.safetensors
    ├── tokenizer.json
    └── tokenizer_config.json