SteelStorage/AbL3In-15B

SteelStorage/AbL3In-15B is a 15 billion parameter instruction-tuned Llama-3 based model, created by SteelStorage, developed through a passthrough merge method utilizing a zeroing technique. This experimental model is derived from failspy/Meta-Llama-3-8B-Instruct-abliterated-v3, focusing on exploring specific merge strategies. It achieves an average score of 67.46 on the Open LLM Leaderboard, with notable performance in reasoning and common sense benchmarks.