Model Overview

princeton-nlp/Llama-3-Instruct-8B-ORPO is an 8 billion parameter language model derived from the Llama-3 architecture, featuring an 8192 token context length. Developed by princeton-nlp, this model's primary distinction lies in its fine-tuning methodology: it utilizes SimPO (Simple Preference Optimization). SimPO is a novel, reference-free reward approach for preference optimization, as introduced in the research preprint "SimPO: Simple Preference Optimization with a Reference-Free Reward".

Key Characteristics

• Architecture: Llama-3 base model.
• Parameter Count: 8 billion parameters.
• Context Length: Supports an 8192 token context window.
• Fine-tuning Method: Employs SimPO, a preference optimization technique that operates without requiring a reference reward model.

Intended Use

This model serves as an implementation and demonstration of the SimPO fine-tuning method. It is particularly relevant for researchers and developers interested in:

• Exploring alternative preference optimization techniques for large language models.
• Understanding the practical application of reference-free reward methods.
• Evaluating the performance of models fine-tuned with SimPO, as detailed in the associated research paper and repository.