T145/ZEUS-8B-V2-abliterated

T145/ZEUS-8B-V2-abliterated is an 8 billion parameter causal language model derived from T145/ZEUS-8B-V2, featuring a 32768 token context length. This model has undergone a specific 'abliteration' process to reduce its refusal capabilities by modifying specific layers based on calculated refusal directions. It is primarily intended for research into model safety and the effects of targeted intervention on model behavior, rather than general-purpose application.