This update brings support to the latest SOTA LLM model: Llama3!
Full Changelog
New features:
Llama3 family models support!
experimental flash attention support (turn on in Advanced settings), ~10% improvement in loading and response times
added ability to re-order apps
added UI settings to show small app icons
Improvements:
keep screen awake during chatting, PDF processing, and other long-running tasks
diagnostics feature now fixes common errors encountered during LTM ingestion automatically
long running processes such as LTM, Tasks Manager, and PDF Digest now has a special interface which automatically dims your screen for processing during the night
OpenAI and Claude APIs now support long term memory, PDF digest, and the new edit & regen method
added a fallback phrase for Todo app
if you have background inference enabled, task manager tasks will execute in the background as a widget when you click "execute all"
added button in the Background Inference app that kills off any floating widgets to handle rare cases were a widget gets stuck and would not exit until your phone is restarted
added ability to execute a group of tasks with one button
Bug fixes:
fixed bug where chat history sometimes duplicates your messages
fixed bug where toolkit was not working iOS
fixed bug where not all events were being displayed in the Todo app
fixed bug where edited messages were not saving
fixed bug where TTS was crashing the app
Comments