Greytechai/Meta-Llama-3-8B-Instruct-abliterated-v3
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kPublished:Mar 16, 2026License:llama3Architecture:Transformer Warm

Greytechai/Meta-Llama-3-8B-Instruct-abliterated-v3 is an 8 billion parameter instruction-tuned causal language model based on Meta-Llama-3-8B-Instruct. Developed by Greytechai, this model has undergone a unique "abliteration" process using orthogonalization to specifically inhibit refusal behaviors, making it less prone to ethical/safety lecturing while retaining the original model's knowledge and training. It is primarily designed for use cases where a less censored, more direct response is desired, offering a surgical modification without broad behavioral changes typically seen in fine-tuning.

Loading preview...