paulilioaica/Hugo-7B-slerp
paulilioaica/Hugo-7B-slerp is a 7 billion parameter language model merged from Mistral-7B-Instruct-v0.2 and CodeNinja-1.0-OpenChat-7B using the slerp method. This model demonstrates improved performance over its base models in specific benchmarks like ARC, MMLU, and Winogrande, while maintaining a 4096-token context length. It is designed for general conversational tasks with enhanced reasoning capabilities, particularly in areas where its merged components excel.
Loading preview...
Overview
paulilioaica/Hugo-7B-slerp is a 7 billion parameter language model created by merging mistralai/Mistral-7B-Instruct-v0.2 and beowolx/CodeNinja-1.0-OpenChat-7B using the slerp (spherical linear interpolation) merge method via mergekit. This approach combines the strengths of both base models, aiming for a balanced performance across various tasks.
Key Capabilities & Performance
The merge operation has resulted in a model that shows improved performance in several key benchmarks compared to its Mistral-7B-Instruct-v0.2 base model. Specifically, Hugo-7B-slerp achieves:
- 64.51 on ARC (AI2 Reasoning Challenge)
- 62.54 on MMLU (Massive Multitask Language Understanding)
- 80.03 on Winogrande
While its overall average score on the Open LLM Leaderboard is 67.07, it notably surpasses the base Mistral model in these specific reasoning and understanding tasks. The model maintains a 4096-token context length.
When to Use This Model
This model is suitable for general conversational AI applications where a balance of instruction following and reasoning capabilities is desired. Its enhanced performance in specific benchmarks suggests it could be particularly effective for tasks requiring logical deduction and comprehensive understanding, building upon the strong foundation of its Mistral and CodeNinja components.