2026-10-16 17:22 UTC
Using the 4096-token local Apple Intelligence model to parse garage invoices was quite a challenge. Here’s a technical look behind the scenes of Vehicare:
https://medium.com/@pascalbourque/reading-invoices-with-apples-on-device-model-in-4096-tokens-1fbeaea0b839
#IndieDev #AppleIntelligence #llm #AppleDeveloper
Replies (1)
-
@benchsmoke@mastodon.social 2026-10-16 18:52
@bourquep@mastodon.social Letting the layout decide the amounts while the model only points at line numbers is a sensible split when the window is this small. Did it hold up on garage formats you hadn't tuned on, or did new layouts still need rules added?