failspy/Meta-Llama-3-8B-Instruct-abliterated-v3

failspy/Meta-Llama-3-8B-Instruct-abliterated-v3 is an 8 billion parameter instruction-tuned Llama 3 model developed by failspy. It features orthogonally ablated bfloat16 safetensor weights, specifically manipulated to inhibit refusal behaviors without altering other core functionalities. This model is optimized for use cases requiring an uncensored response style while retaining the original Llama 3's knowledge and training.