---
language: en
tags: [autoscientist, adaption-labs, qlora, science, dpo]
base_model: Qwen/Qwen2.5-0.5B-Instruct
datasets: [Rishidar/autoscientist-science-dataset]
---

# AutoScientist Competition — Science Model

Qwen2.5-0.5B-Instruct adapted for **science** via Adaption Labs AutoScientist v5:
4-bit QLoRA SFT (r=32, alpha=64) then DPO (beta=0.1) on chosen/rejected pairs.

- DPO reward accuracy: 0.7272727272727273
- DPO reward margin: 4.214854045347734

Dataset: [Rishidar/autoscientist-science-dataset](https://huggingface.co/datasets/Rishidar/autoscientist-science-dataset).
Also mirrored on Kaggle: rishidard/autoscientist-science-qlora.