A fine-tuned variant of Meta's Llama 3 70B, published by Cognitive Computations (dphn), configured to operate without the typical safety guardrails found in base models. It handles a wide range of text tasks with the capability you'd expect from a 70B parameter model, but its uncensored nature means responses can be more direct and unfiltered than its base counterpart. The trade-off is that it requires more careful deployment consideration.