Model Overview

The rwitz2/go-bruins-v2.1.1 is a 7 billion parameter language model, fine-tuned using DPO (Direct Preference Optimization) on the Intel/orca_dpo_pairs dataset. It is based on the jan-hq/trinity-v1 architecture. At its release, this model achieved a notable position as the #1 model on certain leaderboards, indicating strong performance in its category.

Key Characteristics

• DPO Training: Utilizes Direct Preference Optimization for enhanced instruction following and response generation.
• Leaderboard Performance: Achieved top rankings on leaderboards shortly after its release (December 2023).
• Derived from jan-hq/trinity-v1: Builds upon an established base model for its capabilities.
• Contains Traces of Cybertron-2: The model incorporates elements from the fblgit/una-cybertron-7b-v2-bf16 project, as acknowledged by the creator.

Important Considerations

The creator has noted potential contamination in some of the datasets used for fine-tuning and is actively working to address this in future iterations. Users should be aware of this ongoing refinement process.

Good For

• General language generation tasks where DPO-tuned models excel.
• Applications requiring strong instruction following based on preference optimization.
• Experimentation with models that have demonstrated high leaderboard performance.