Model Overview

jan-hq/trinity-v1 is a 7 billion parameter language model developed by Jan, utilizing the Slerp merge method. It integrates two high-performing models from the OpenLLM Leaderboard as of December 14th: viethq188/LeoScorpius-7B-Chat-DPO and GreenNode/GreenNodeLM-7B-v1olet, with the latter serving as its base model. This model is a test project for merging techniques, aiming to combine the strengths of its constituent models.

Key Capabilities & Performance

Trinity-v1 demonstrates solid performance across a range of benchmarks, as evaluated on the OpenLLM Leaderboard. Its average score is 74.8, with specific results including:

• ARC (25-shot): 72.27
• HellaSwag (10-shot): 88.36
• MMLU (5-shot): 65.2
• TruthfulQA (0-shot): 69.31
• Winogrande (5-shot): 82
• GSM8K (5-shot): 71.65

Usage and Development

This model is optimized for use with a specific prompt template:

{system_message}
### Instruction:
{prompt}

### Response:

It can be run locally using Jan Desktop, an open-source, offline-first ChatGPT alternative that offers an OpenAI-compatible local server. The development of this model leverages tools like mergekit and draws inspiration from techniques like DARE and SLERP for model merging.