0.3.49: threesome, grok, bugfix
Added 2024-12-13 13:08:40 +0000 UTCโจ a feature with two characters on the screen added to the sandbox *
โจ support for the grok-beta model (the one from E Musk) **
โจ a button for forcibly returning Dina to the screen
๐ a critical bug with custom contexts for the female hero fixed, now everything works including dnd mode
๐ maximum temperature reduced to 1.8
๐ gemini-1.0-pro and gpt-3.5-turbo-1106 models removed
๐ dark theme fixed
๐ apk: less screen dimming while typing
๐ gemini: optimized the work of bots responsible for changing locations and saving history; the location-bot now works correctly, always creates a new empty session and only looks at the last 2 messages, not the entire history; the history-saving bot no longer has redundant duplicated text. especially relevant for long gaming sessions
๐ in the prologue dream, the embarrassing (poor-quality) videos with dina and lora have been removed, replaced with images
๐ music files re-encoded into ogg (same quality but half the size)
๐ all webm videos slightly compressed without visible quality loss
๐๏ธ person and base passwords changed
*invocation in the top-left corner, can be called by text as well, but the menu is more reliable. it works like this: if two characters are on the screen, a hidden bot decides before each response who should speak, switching from one character to the other along with all their custom contexts
** grok - interesting, has a free $25 trial, seems not to censor at all; be cautious with temperature, above 1 it starts hallucinating; gets buggy, repeats phrases in prolonged conversations
the main problem with multiple characters is that almost all language models only have a marker with two options - either the user or the assistant/model. that is, there is no separate field to indicate a character, so we have to ask them to add "(Name):" to differentiate speech. haven't tested with local models, but it should be fine
have a nice weekend )
