Stopwolf/DistilabelCerberus-7B-slerp
DistilabelCerberus-7B-slerp is a 7 billion parameter language model created by Stopwolf, formed by merging dvilasuero/DistilabelBeagle14-7B and teknium/OpenHermes-2.5-Mistral-7B using a slerp merge method. This model demonstrates improved performance across reasoning and common sense benchmarks, including ARC-C, HellaSwag, and GSM8K, compared to its base models. It is optimized for general language understanding and generation tasks, leveraging the strengths of its constituent models.
Loading preview...
DistilabelCerberus-7B-slerp Overview
DistilabelCerberus-7B-slerp is a 7 billion parameter language model developed by Stopwolf, created through a 'slerp' merge of two distinct models: dvilasuero/DistilabelBeagle14-7B and teknium/OpenHermes-2.5-Mistral-7B. This merging technique combines the strengths of both base models to achieve enhanced performance.
Key Capabilities & Performance
This merged model shows notable improvements in several key benchmarks:
- Reasoning: Achieves 65.44 on ARC-C, surpassing OpenHermes-2.5-Mistral-7B's 61.26.
- Common Sense: Scores 69.29 on HellaSwag and 79.48 on Winogrande, indicating strong common sense reasoning.
- Mathematical Reasoning: Demonstrates significant improvement on GSM8K with a score of 69.82, compared to OpenHermes-2.5-Mistral-7B's 26.08.
- Overall Evaluation: The model achieves an average score of 71.56 on the Open LLM Leaderboard, with specific metrics including 68.17 on AI2 Reasoning Challenge and 64.20 on MMLU.
When to Use This Model
DistilabelCerberus-7B-slerp is a strong candidate for applications requiring robust general language understanding and generation, particularly where improved reasoning and mathematical capabilities are beneficial. Its balanced performance across various benchmarks makes it suitable for a wide range of tasks, offering a more capable alternative to its individual base models.