nbeerbower/llama-3-spicy-abliterated-stella-8B

Model Overview

nbeerbower/llama-3-spicy-abliterated-stella-8B is an 8 billion parameter language model created by nbeerbower. It was developed using the Model Stock merge method, with nbeerbower/llama-3-spicy-8B serving as the base model. This merge incorporates three distinct models:

• nbeerbower/llama-3-stella-8B
• saishf/Aura-Uncensored-OAS-8B-L3
• cognitivecomputations/Llama-3-8B-Instruct-abliterated-v2

Performance Benchmarks

Evaluated on the Open LLM Leaderboard, this model demonstrates a balanced performance across various tasks, achieving an average score of 69.27. Key benchmark results include:

• AI2 Reasoning Challenge (25-Shot): 65.19
• HellaSwag (10-Shot): 81.98
• MMLU (5-Shot): 67.88
• TruthfulQA (0-shot): 53.65
• Winogrande (5-shot): 76.48
• GSM8k (5-shot): 70.43

Key Capabilities

• Reasoning: Strong performance in reasoning tasks, as indicated by its AI2 Reasoning Challenge and GSM8k scores.
• Common Sense: Exhibits good common sense understanding, reflected in its HellaSwag and Winogrande results.
• Language Understanding: Capable of general language understanding and instruction following, supported by its MMLU score.

When to Use This Model

This model is suitable for applications requiring a general-purpose 8B parameter LLM with a balanced performance profile across various benchmarks. Its merged architecture suggests a blend of capabilities from its constituent models, making it a versatile choice for tasks involving reasoning, common sense, and instruction-following.