assn2-simpo-qwen2.5-1.5b-lora

SimPO LoRA adapter for Qwen/Qwen2.5-1.5B-Instruct.

This repository contains a LoRA adapter trained for CAS4133 Assignment 2.

Base model

Qwen/Qwen2.5-1.5B-Instruct

Evaluation accuracy on 100 held-out GSM8K-style examples

Model	Accuracy
Base	0.69
SFT	0.58
DPO	0.57
SimPO	0.58

Dataset

https://huggingface.co/datasets/Riacrdo/assn2-preference-dataset

Note

The full merged model was not uploaded due to network instability. This adapter can be loaded with the base model.