Nitral-Archive/Hex-Macaroniac-7b
Nitral-Archive/Hex-Macaroniac-7b is a 7 billion parameter language model created by Nitral-Archive, formed by merging jeiku/SpaghettiOs_7B and jeiku/Nitrals_Monster_7B. This model leverages a slerp merge method to combine the strengths of its base models, achieving an average score of 66.64 on the Open LLM Leaderboard. It demonstrates balanced performance across various benchmarks, including reasoning, common sense, and language understanding tasks, making it suitable for general-purpose applications requiring a compact yet capable model.
Loading preview...
Hex-Macaroniac-7b: A Merged 7B Language Model
Hex-Macaroniac-7b is a 7 billion parameter model developed by Nitral-Archive, created through a strategic merge of two existing models: jeiku/SpaghettiOs_7B and jeiku/Nitrals_Monster_7B.
Merge Configuration
This model was produced using a slerp (spherical linear interpolation) merge method. The configuration specifically weighted different layers and components of the source models, with varying t values applied to self-attention and MLP layers to optimize the blend.
Performance Highlights
Evaluated on the Open LLM Leaderboard, Hex-Macaroniac-7b achieved an overall average score of 66.64. Key benchmark results include:
- AI2 Reasoning Challenge (25-Shot): 65.53
- HellaSwag (10-Shot): 84.68
- MMLU (5-Shot): 62.43
- Winogrande (5-Shot): 78.30
- TruthfulQA (0-shot): 55.93
- GSM8k (5-shot): 52.99
Use Cases
Given its balanced performance across a range of benchmarks, Hex-Macaroniac-7b is well-suited for general language understanding and generation tasks where a 7B parameter model is appropriate. Its merge-based origin suggests a blend of capabilities from its constituent models, making it a versatile option for various applications.