nanoGPT - is anyone playing with it?
Simon, a dude I follow on Twitter custom trained nanoGPT (a 4-layers model) to write in a Shakespeare style.
On his Mac. :)
Is it just a toy or you can do some useful stuff?
Trending on Indie Hackers
The fastest way to turn a slide deck into a 24/7 funnel
I tried it with one of the GPT2 models on a small private project to do word synonyms, but it wasn't super for it. In comparison to the GPT3 api from OpenAI it's lacking the possibility to use semantic styles (positive, informal, catchy synonym etc.).
I think it would be possible to use it to train specific types of writing styles with it, and it is made for being used with high grade consumer machines as you mention. The stuff I've done with Stable Diffusion or other stuff usually requires A100 GPU's and stuff.
The shakespeare example anyone can try, it's even documented in the repo: https://github.com/karpathy/nanoGPT
thanks, I kind of like the fact there is a bunch of models, simple and super complex so you can match a tool to a specific need; also environmental concerns...
I didn't know about this. Thanks for sharing!
Definitely think it can be useful. But it's a bad generalist AI and only really good for very specific use cases. Especially considering these are 124m params, which is quite low.
Hoping to get this on Evoke. We just have stable diffusion for now, but are expanding into other AI models, so nanogpt definitely seems interesting.
Also have an active AI discord if you're interested.
thanks, discord joined.