Releases: TavernAI/TavernAI
@1.5.4
@1.5.3
2024 'rollup' release.
Time to nudge the version up, so the update alert goes out... Happy New Year!
What's Changed
- Explicit address binding by @ExZyle in #248
- Brought back the option to choose gpt-3.5-turbo-0301 by @FunkEngine2023 in #249
New Contributors
- @ExZyle made their first contribution in #248
- @FunkEngine2023 made their first contribution in #249
Full Changelog: https://github.com/TavernAI/TavernAI/compare/@1.5.2...@1.5.3
@1.5.2
@1.5.1
###Update 1.5.1###
Improved chat book display
Added the function of changing the chat name in Chat Book
Added the depth setting for system prompt and jailbreak prompt
Text generation web UI supporting
Implemented selected characters in an existing room (ufeike)
Implemented selected character removal in an existing room (ufeike)
The context for KoboldAI has been increased to 8192 and manual input has been added
Added support for importing v2 version of character cards
Added supporting of Kayra NovelAI model
Fixed notable bugs
Fixed a bug when the System Prompt preset was reset when changing the character
Fix: The disappearance of the text for the story mod when entering a text through the field for sending messages
Fixed a bug when the "Chat Book" disappeared when switching between rooms and regular chat
Fixed a bug when the send text field did not return its original size when sending a message
Autorun fix
Another lots of minor fixes
@1.5.0
TavernAI 1.5.0 Stories
-
Story mode
New mode for writing stories and novellas. Instead of writing responses for chat, the AI will continue the already written text. The convenience is that you can switch to chat / story mode in real time.
In addition to Davinci, three additional models for writing stories have been added: Ada, Babbage, Curie. (These four models are uncensored)
Added a new default character card as an example for story mode -
Character rooms support (ufeike)
Character rooms allow users to use a chat with multiple characters simultaneously, as well as a chat between AI characters without user involvement. -
Improved work with proxies, added proxy support with passwords (Aibo)
-
Reverse Proxies have been moved to a separate api
The available proxy models are now loaded from metadata and appear in the list
The maximum context for the proxy has been increased to 100,000.
Also added manual introduction of context size -
Added the ability to save and select presets for the system prompt and jailbreak
-
Added dynamic resizing of the text field to send a message in a classic theme
-
Added gpt-3.5 16k model for OpenAI
-
For main page of TavernAI the summer style is used
-
Fixed notable bugs:
Duplicate categories when switching nsfw toggle in the charaCloud
Fixed styles and markup for world info
Fixed an autoconnect that tried to connect to all available api at the same time
Fixed additional templates
Too long character name is now shortened in the upper right menu, and does not get out of the screen
Lots of minor fixes
@1.4.1
###Update 1.4.1###
- Categories autocomplete for online character editor
- Folders for characters (Aibo)
- Improved mobile layout on all themes (weebcouncil)
- Importing characters by dragging an avatar onto the page
- Chats export/delete
- Added NSFW filter toggle for showing characters on/off
- Proxy support for OpenAI
- NovelAI Clio model
- Top categories on charaCloud main page
- Fixed showing random characters
- Changing the user name without reloading the page
@1.4.0
- The ability to upload characters and create own collections online via TavernAI interface has been added. New online features:
- Special Editor for uploading and updating online cards
- Categories for character cards
- Online user profiles
- Changing an user avatar for online profiles
- The ability to upload characters both from TavernAI library and by selecting a file
- Editing and deleting already uploaded characters
-
Memory Scroll (Aibo)
A function for creating notes that are taken into account by AI. These can be both events that occurred in the chat, which the AI must remember, and any other information. Unique for each chat. -
OpenAi Jailbreak
New Jailbreak system for OpenAI has been added -
World Info (Aibo)
The classic World Info feature has been added -
Horde (Aibo)
Horde support has been added -
User avatars
Added the ability to upload and delete user avatars via TavernAI interface -
TavernAI Code
Added a new message processing system with convenient code display -
Fixed a bug non-english characters in webp
Now the webp format accepts any characters and any languages -
Many other minor fixes and improvements
@1.3.1
@1.3.1
- Added autoconnect checkbox
- Advance message edit options (clone message, delete message, move up/down, character name selector)
- Singleline mode toggle for KoboldAI
- Fixed a bug when the character name was not added for KoboldAI api
- OpenAI "system prompt" edit function for GPT-Turbo models
- Swipes with the arrow keys
- Partial code refactoring
- Added a start script for linux
- Updated GPT tokenizer
- Security Update (Set default server bind address to loopback)
- Removed background blur which slowed down some systems
- CharaCloud toggle in settings in menu (Accordingly, the same setting is removed from the config.conf)
- Fixes to character info breaking due to "smart" quotes
- Fix bug: If you used swipes, while there is any text in the text field an extra message with an ellipsis is added
- Fix: infinite loop when looking for message root, possibility to edit multiple messages
@1.3.0
@1.3.0
-
charaCloud - online characters database has been launched
charaCloud interface is now integrated into TavernAI, allowing for character search and selection (the first and second stage of the roadmap).
charaCloud can be enabled/disabled through the config.conf -
Settings
The settings have been renamed to Master Settings and are now a separate dynamic menu with options for both TavernAI and each selected API.
Repetition Penalty Range and Anchors Order has been moved from the side panel to Master Settings
All types of samples are now editable for each API.
Multi-generation option enable/disable
Keep dialogue examples option enable/disable
Free name mode enable/disable (Mode for narrative models)
For KoboldAI Context Size added a lock option that allows the setting to remain fixed when switching presets.
Pygmalion formatting option (Auto, Enable for all models, Disable for all models)
-
Work has been carried out to optimize character cards. A new .webp storage format is available (the quality is the same, the size is 8 times smaller). This reduces the load on both charaCloud and the local computer running TavernAI. The old png format is still available, and the conversion process happens automatically. The format can be switched in the config.conf
-
OpenAI API
Support for the OpenAI API has been added, including five main models (gpt-3.5-turbo, gpt-3.5-turbo-0301, text-davinci-003, text-davinci-002, code-davinci-002) -
Swipes
Added swipes for messages
The option of disabling/enabling swipes if necessary -
Designs
The CSS styles have been completely overhauled into a modular structure. The designs can be switched in the left menu where the background selection is located. To add a new design, simply add a CSS file to public\designs, and it will automatically appear in the left menu. The new design inherits the classic design and can simply override or disable inherited properties in the new file.
New designs have been added:
Amber - by Luizsan
Lite (cai like design) - Developed jointly with RossAscends
The auto-resize field feature (developed by RossAscends) has been added and can be enabled or disabled for custom designs through the styles (by default, enabled for Lite design).
- Numerous minor fixes and improvements.
@1.2.8
@1.2.8
- Security update: GHSA-25wp-82wc-xchj
- Fixed several bugs related to character import: now all imported characters are added to the top of the list.
- Fixed a bug that duplicated character cards if the character name and map name did not match.
- Made minor fixes for prompt formatting.
- Added the ability to enable/disable TavernAI autostart in the browser through the configuration file.
- Adapted for Pygmalion Dev.