simpo-training
orchestra-research892 installs
Align language models without a reference model.
What it does
Guides language-model preference training with SimPO without maintaining a reference model.
When to use it
Use when preference pairs are available and simpler, reference-free training is wanted, particularly for limited-compute, single-node work.
How to use it
Give it a language-model alignment task; it provides training configuration and launch steps that produce a trained model in the project.
Uses
Latest review
What you need · 8
Uses conda to create and activate the training environment.
The setup specifies Python 3.10 in the training environment.
Access · 6
This skill
https://github.com/huggingface/alignment-handbook.git
Read
Downloads the training repository
Hugging Face
Read
Loads model and preference data