Model Profile

FamilyCognitive Computations / Meta Llama 3
TypeBase Model
Parameters8B
Ollamadolphin-llama3:8b
Platform
Last seen

Purpose

Evaluate model behaviour with reduced safety guardrails

About This Model

Dolphin Llama 3 8B is an uncensored fine-tune of Meta Llama 3 produced by Eric Hartford. It is used on OMEGA Core to evaluate AI behaviour in the absence of standard safety training, providing a reference point for understanding what guardrails actually prevent and how they can be studied in a controlled environment. It also serves as the base model for omega-therapist, making the interaction between an uncensored foundation and a role-based system prompt a deliberate research variable.

Evaluation Findings

Findings will be documented here as evaluation progresses. This is a skeleton page — content is added as research is conducted.