Where we’re at with self-hosted AI today?

I have lost track of what has been happening in GPT space.

Last I heard about was LLAMA leaks.

Are there any further developments in this space? I’m presuming there must be loads and I’d really like to try out some models on my new 4090.

Thanks!

1 comment

[ 2.4 ms ] story [ 10.2 ms ] thread

smoldesu 3y ago ↗

There are a lot of options. Right now I'm using WizardVicunaLM to great success: https://github.com/melodysdreamj/WizardVicunaLM

It combines the uncensored WizardLM data with the Vicuna tuning to create a surprisingly high-performance model. If the chart on their GitHub page is to be believed, their model approaches GPT-3.5 performance.