| GitHub | https://github.com/jankais3r |
| https://twitter.com/jankais3r |
| GitHub | https://github.com/jankais3r |
| https://twitter.com/jankais3r |
The on-device LLM model bundled in iOS 18.1 Beta 1 seems to have a data cut-off date some time in 2023, however, for some categories of questions it answers with 2022 or even 2021 data points.
Answering questions is, of course, not what the model was designed for. It’s supposed to rewrite your emails and messages. But, with a special prompt, you can turn it into a very primitive chat bot 🙂
Credits for the discovery of control tokens (extracted from macOS) go to: https://gist.github.com/EvanZhouDev/1a5d3e3705612f56b6aaa09fe862ec47
Half of GitHub is struggling to run the 7B #LLaMA model on their desktop GPUs, meanwhile MacBooks can run the 13B model without breaking a sweat thanks to their Unified Memory Architecture.
Code here: https://github.com/jankais3r/LLaMA_MPS
Can’t believe this passed the TestFlight beta review 😅
TestFlight link: https://testflight.apple.com/join/bJyw9fog
GitHub repo: https://github.com/jankais3r/quake_watch