DareBeagle-7B Overview

DareBeagle-7B is a 7 billion parameter language model developed by shadowml. It is a product of a strategic merge between two base models: mlabonne/NeuralBeagle14-7B and mlabonne/NeuralDaredevil-7B. This merge was executed using the slerp (spherical linear interpolation) method, specifically configured to blend the self-attention and MLP layers of the constituent models.

Key Capabilities & Performance

DareBeagle-7B exhibits strong performance across a range of benchmarks, as evaluated on the Open LLM Leaderboard. Its average score is 74.58, indicating solid general-purpose reasoning and language understanding. Specific benchmark results include:

• AI2 Reasoning Challenge (25-Shot): 71.67
• HellaSwag (10-Shot): 88.01
• MMLU (5-Shot): 65.03
• TruthfulQA (0-shot): 68.98
• Winogrande (5-shot): 82.32
• GSM8k (5-shot): 71.49

Good For

• General-purpose text generation: Its balanced performance across various benchmarks makes it suitable for a wide array of language tasks.
• Reasoning tasks: Demonstrated proficiency in AI2 Reasoning Challenge and GSM8k suggests capabilities in logical inference and problem-solving.
• Applications requiring robust understanding: High scores on HellaSwag and Winogrande indicate strong common sense and contextual understanding.