Today, the non-profit Consumer Wellness Center Data Science Division is announcing the free release of two new AI Large Language Models, available now at no cost at Brighteon.AI.

The two new models are:

Neo-Phi-2-E3-1-7B-V0-1-1-Q8-Apr-24 - Based on Microsoft's Phi-2 1.7 billion parameter model, we have conducted 3 epochs of training on a very large data set (see below), vastly improving the model's score on our "LLM Reality Check" scoring system which consists of 100 questions covering medicine, science, history, nutrition, finance and more. Phi-2 scores 54 points out of 100 as the base model. After our training, we improved the score to 79 points out of 100. This is a very fast, computationally-efficient model that is suitable for smaller computers or even laptops.

Neo-Dolphin-Mistral-7B-E3-V0-1-1-Q8-Apr-24 - Based on Mistral's 7 billion parameter model, modified into "uncensored" behavior by Eric Hartford, we applied 3 epochs of additional training on our procured data set. Because this is a much larger model, the training has not yet achieved the level of saturation that we are aiming for. We do not yet have a score on this model's output, but it won't be impressive yet. More training is needed (and is being pursued).