This version brings better sentiment detection to Layla! In the form of a new app: Character Expressions!
Upload up to 28 unique expressions for your characters. During your chat, Layla will use real-time sentiment detection to update the chat background according to the current emotion of your conversation.
Additionally, the sentiment detection is applied to auxiliary features such as Layla's Toolbox to provide better detection of your intent. For example, Layla will now only search the web if your detected emotion is curious. This should eliminate a lot of the false positives during tool use.
Quick Actions
This update continues to expand on the OS integration in Android and Siri Shortcuts. In the Quick Actions app, you can customise the prompts and instructions used when "share to Layla" is triggered. This gives you much more flexibility in controlling what instructions to give your characters when text is shared to Layla.
You can attach your instructions to a specific character for even more flexibility.
The video below demos how a custom instruction can be set to receive bug reports from Discord and remind me to fix them tomorrow!
Your imagination's the limit on what instructions you put for your characters!
Executorch update
The Executorch app is further integrated into Layla. You can now load custom models into Executorch. You can do this the same way you do for GGUF files: go to the RMI (Raw Model Instructions) app and load a PTE file downloaded from Hugging Face.
My repo contains a variety of different Executorch compilation you can try out: https://huggingface.co/l3utterfly
Model requests are welcome! If you find a fine-tuned Llama3 model that you would like to use with Executorch, feel free to reach out to me on Discord and I will compile them for you.
Full Changelog
New features:
Character Expressions app: choose a unique image for 28 unique emotions your character feels during chatting!
Quick Actions app: allows you to customise instructions for Siri shortcut and Android OS integration
Improvements:
Executorch now uses RMI app to manage models
streamlined task manager task management
task manager now schedules tasks much more consistently
task manager now supports OpenAI API backends for long term memory ingestion, dream scheduling, and pdf digest
improved handling of stop button during chat
advanced settings default preset now resets everything
Bug fixes:
fixed bug where task manage queued tasks were not displaying properly
fixed bug where background inference was not running properly
Komentar