Model Overview

Azazelle/smol_bruin-7b is a 7 billion parameter language model developed by Azazelle. This model is a Slerp merge of two distinct models: rwitz/go-bruins-v2 and rishiraj/smol-7b, both based on the mistralai/Mistral-7B-v0.1 architecture. The merge process utilizes specific t values for self-attention and MLP layers, indicating a fine-tuned combination of their respective strengths.

Key Capabilities

• General Reasoning: Achieves a score of 67.58 on the AI2 Reasoning Challenge (25-Shot).
• Common Sense Reasoning: Scores 86.48 on HellaSwag (10-Shot) and 81.14 on Winogrande (5-shot).
• Mathematical Reasoning: Demonstrates capability with a 70.43 score on GSM8k (5-shot).
• Knowledge & Comprehension: Attains 65.05 on MMLU (5-Shot) and 55.65 on TruthfulQA (0-shot).

Performance Highlights

Evaluated on the Open LLM Leaderboard, smol_bruin-7b achieved an average score of 71.05. This performance positions it as a capable model for various general-purpose applications requiring strong reasoning and language understanding within its 7 billion parameter class.