Model Overview

This model, princeton-nlp/Mistral-7B-Base-SFT-SLiC-HF, is a 7 billion parameter language model built upon the Mistral architecture. It was developed by Princeton NLP as part of their research into preference optimization techniques.

Key Characteristics

• Architecture: Based on the Mistral-7B-Base model.
• Fine-tuning Method: Utilizes the SLiC (Supervised Learning with Contrastive Loss) method, as detailed in the associated research paper.
• Research Origin: This model is a direct output of the preprint SimPO: Simple Preference Optimization with a Reference-Free Reward, which focuses on developing simple, reference-free reward models for preference optimization.
• Context Length: Supports a context window of 4096 tokens.

Intended Use Cases

This model is suitable for a variety of general natural language processing tasks, particularly those where the benefits of SLiC-based fine-tuning for improved preference alignment are desired. Developers interested in exploring models derived from advanced preference optimization research will find this model relevant.