metadata
language: en
tags:
- autoscientist
- adaption-labs
- qlora
- legal
- fine-tuned
base_model: Qwen/Qwen2.5-0.5B-Instruct
datasets:
- Rishidar/autoscientist-competition-datasets
AutoScientist Competition — Legal Model
Fine-tuned from Qwen/Qwen2.5-0.5B-Instruct on the legal adapted dataset produced by Adaption Labs AutoScientist Challenge.
Training
- Method: QLoRA (4-bit NF4, r=32, alpha=64)
- Epochs: 3
- Learning rate: 0.0002
- Dataset quality: Grade A (Adaption Labs evaluation)
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("Rishidar/autoscientist-legal-qlora")
tokenizer = AutoTokenizer.from_pretrained("Rishidar/autoscientist-legal-qlora")