Nostr Web Client

Playing with AI again. Gpt4all has become fast on M2 chip, cool new civicai finetuned models of Stable diffusion, downloading new quantized falcon models to try with a fork of llama.cpp that should support the architecture.

You miss two weeks and there's so much to try.

https://vimeo.com/837394279

Shawn 2y ago

Damnit, man. I’m up for a new laptop, and I’m trying to tell myself and Air enough. You’re going to make me buy a Pro.

Also, this link is broken.

Reply to this note

Please Login to reply.

Discussion

Juraj 2y ago

Oops.

Air should be also ok for smaller models, they have M2 chip. The main limitation is memory. 40G model in full fp16 precision needs 80GB, less than 40GB in q5.

This is 13B model in q5, so if you have 32GB macbook air, you should be fine. Even iPhones and iPads can technically run these models.

https://vimeo.com/837394279/ee5f723467

Shawn 2y ago

Thank you for this!