arlineka/Brunhilde-13b-v1
arlineka/Brunhilde-13b-v1 is a 13 billion parameter language model created by arlineka, formed by merging Gryphe/MythoMax-L2-13b and Undi95/ReMM-SLERP-L2-13B. This model demonstrates a strong average performance of 57.88 on the Open LLM Leaderboard, with notable scores in reasoning and common sense benchmarks. It is designed for general-purpose text generation and understanding tasks, leveraging the combined strengths of its constituent models.
Loading preview...
Brunhilde-13b-v1: A Merged 13B Language Model
arlineka/Brunhilde-13b-v1 is a 13 billion parameter language model developed by arlineka, created through a strategic merge of two established models: Gryphe/MythoMax-L2-13b and Undi95/ReMM-SLERP-L2-13B. This merging approach aims to combine the strengths of its base models to enhance overall performance across various language understanding and generation tasks.
Key Capabilities & Performance
Brunhilde-13b-v1 has been evaluated on the Open LLM Leaderboard, achieving a solid average score of 57.88. Its performance highlights include:
- AI2 Reasoning Challenge (25-Shot): 61.09
- HellaSwag (10-Shot): 83.58
- MMLU (5-Shot): 55.32
- Winogrande (5-Shot): 75.22
These scores indicate a robust capability in reasoning, common sense, and general knowledge tasks. While its GSM8k (mathematical reasoning) score is 20.09, it shows strong aptitude in other critical areas.
Good For
- General Text Generation: Suitable for a wide range of creative and informative text generation needs.
- Reasoning Tasks: Excels in tasks requiring logical deduction and understanding, as evidenced by its AI2 Reasoning Challenge score.
- Common Sense Understanding: Strong performance on HellaSwag and Winogrande suggests good common sense reasoning abilities.
Developers can integrate this model using the Hugging Face transformers library, with provided Python code for quick setup and inference.