J.O.S.I.E.v4o

is my ongoing project on creating a fully multimodal modal, that understand images, videos, thermal images, depth, and audio. J.O.S.I.E. is not only a conversational general purpose AI assistant, it’s also build to manage a smart home, smart and autonomously. exactly like J.A.R.V.I.S.

This model is trained on top if Llama3 8B

This model is in it’s first stage with is the basic instruction fine-tuning stage.

the used Prompt template is inspired my the ChatML template but with adjustments:

At first the main user "Gökdeniz Gülmez" can be changed and updated with your name. it will be further trained to feature authorized user "{name}" just be a guest user without rghts to control or access smart home informations. unauthorized user "name if possible else unknown" this user will not have any access to J.O.S.I.E.’s abilities. and will be redirected to the main, user or other security measures will be taken.

<|begin_of_text|>system
{{ .System }}<|end_of_text|>
<|begin_of_text|>main user "Gökdeniz Gülmez"
{{ .Prompt }}<|end_of_text|>
<|begin_of_text|>josie
{{ .Response }}<|end_of_text|>