Layla v6.2.0 has been published!

Roleplay Update

The Roleplay mini-app has been reimagined in this update! It now offers much more flexibility.

• You can control which character speaks next in the scene, this gives you much more control in how the scenario unfolds

• You can also speak as any character, giving you a chance to keep the story in the direction you want, or reign the story in in case the LLM goes off the rails

• You are always able to speak as the "Narrator". You can use this to advance the scene forward, give new challenges for your characters, and more! It's all up to your imagination

• the "random" button still works as before, randomly choosing the next character to speak

Tip: if you want to participate in the scenario, you can create a custom character representing yourself and add it as a normal participant

Model Benchmarks

Model benchmarks now allow you to share your results anonymously on the benchmark site. You can also use the site to see what should be the expected performance of your phone when running local LLMs.

Please share your benchmarks! It will help new users immensely.

Full Changelog

New features:

• Roleplay app gives much more flexibility - choose who talks next, speak as any of the participants

• Roleplay now supports speaking out load responses with the individual participants voice

• Roleplay app now supports using AI to generate your scenarios and images

  
  

• view publicly submitted data for Mobile Model Benchmarks

• you can submit your own benchmarks by running models through the Model Benchmark mini-app

• in the future, this data will be used to recommend the best models for your phone

  
  

  (All data submission is voluntary, NOT automatically collected.)

Improvements:

• added ability to search for supported models in your OpenAI API endpoint inference settings

• Diagnostics now show a breakdown of storage used by Layla

• added ability to stop LTM from auto-collecting memories, allowing more fine-grained control

Bug fixes:

• fixed a few bugs in Dreams scheduling and improved display of already scheduled Dreams

• fixed a bug where context prefix inference settings were not getting saved

• fixed a bug where character search in character selection switching back to the first page

• fixed a bug where character expressions cannot be set

• fixed a bug where enter key does not clear the sent message