Disclaimer: This release is intended to test device compatibility, you may crash on boot.
This update introduces hexagon kernels for llama.cpp, however device compatibility may be inconsistent.
If you are crashing, please submit a report here with your device name + SOC model:
If you have debugging tools, logs would also be appreciated!
Features:
- updated llama.cpp
- Take photo as input
Changes:
- Updated PNG parser to be 100x faster
- Debounced tokenizer in UIs
- Added parser for reasoning
- Initial theme is now based on system theme
Fixes:
- Notifications causing a softlock
- Tokenizer cache never being used
- Cancelling readers safely
- Missing paste function for text bottom sheets
- Persist taps in the chats drawer
- Stop sequences not being replaced with macros
- Removed Presence Penalty from Google AI
Dev:
- Integrated Nitro modules
- Updated to SDK 54
- Updated MMKV
- Updated to new expo-filesystem