SillyTavern

LLM Frontend for Power Users.

AGPL-3.0 License

Downloads
374
Stars
6.5K
View Code on GitHub Visit Website
Ecosystems: JavaScript

Bot releases are hidden (Show)

SillyTavern - 1.12.1 Latest Release

Published by Cohee1207 4 months ago

SillyTavern 1.12.1

General

  • Node 18 or later is now required to run SillyTavern.
  • The latest tag for GHCR containers now points to the latest release branch push.

Backends

  • Text Generation WebUI: added DRY sampling controls.
  • TabbyAPI: added speculative ngram, skew sampling, and repetition decay controls. Now supports multi-swipe mode.
  • KoboldCpp: added repetition penalty slope control.
  • Google MakerSuite: added custom API URL control. Enabled image inlining for Gemini Flash.
  • Added new models for Cohere and MistralAI. Removed deprecated models from Perplexity.

Improvements

  • Improved sanitation of message-embedded styles.
  • Setting and chat backups are now created in a user-specific folder.
  • Character Cards: first message content is no longer required.
  • Added import of V3 cards characters (PNG, JSON, CHARX). The internal format is still V2.
  • Added characters import from RisuRealm via the external content import dialog.
  • MovingUI now tries to automatically adjust to the resized window instead of resetting to the default position.
  • Markdown: simplified requirements for sublist nesting, can now use any number of spaces. Enabled markdown strikethrough syntax.
  • Macros: {{pick}} and {{random}} can escape commas with a backslash if using a comma as a separator.
  • Data Bank: added bulk edit mode. Markdown imports now preserve formatting.
  • Personas: added in-chat position for descriptions.
  • World Info: restyled global activation settings. Check for world names overlap on import or creation.
  • Preserve itemized prompts history for chat branches and checkpoints.
  • Added a framework for extension authors to use LLM function calling. See the documentation: Function Calling.
  • Various localization updates.

Extensions

  • Added new extensions to the assets list: Screen Share and Prompt Inspector.
  • Image Generation: added Seed and CLIP skip controls for supported backends. Added decrisper control for NovelAI and upscale control for DrawThings. Added scheduler selector for AUTO1111.
  • Image Generation: square brackets, curly braces, and underlines are no longer removed from generated prompts.
  • Image Generation: character-specific prompt prefixes can be saved to shareable card data.
  • Vector Storage: added Ollama and llama.cpp as embedding sources. Added score threshold and chunk overlap settings. Corrupted vector indices are now automatically regenerated.
  • TTS: added Azure TTS and GPT-SoVits Inference providers.
  • Speech Recognition: added KoboldCpp provider. Added hotkey/push-to-talk and voice activity detection modes.
  • Web Search: added regex-activated search query triggers.
  • Top Bar: added quick past chat selection dropdown.

STscript

  • Added a set of commands for Data Bank management. Refer to the in-app documentation for more information.
  • Added a set of commands for tag management.
  • Added ephemeral option for /inject command to set a one-time prompt.
  • /api commands returns the current API name if no arguments are provided.
  • Added /translate command to translate arbitrary text.
  • Added /rename-char command to rename a current character.
  • /flushinject command can now clear a single prompt by its ID. Pass no arguments to flush all prompts.
  • Added /yt-script command for YouTube video script downloading.
  • Updated /ask command to use a new syntax similar to /sendas.

Bug Fixes

  • Fixed chat translations not applying on message edit.
  • Fixed click-to-edit triggering in single document mode if selecting message text.
  • Fixed Chess comment prompt getting stuck in the prompt injects list.
  • Fixed interaction of zoomed avatars and moving UI.
  • Fixed interaction between WI regex keys and global case-insensitive mode.
  • Fixed punycode module deprecation warning.
  • Fixed Image Generation interactive mode trigger being too loose.
  • Fixed deletion of corrupted WI files.
  • Fixed Kobold Horde API domains.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.12.0...1.12.1

SillyTavern - Docker release fix

Published by Cohee1207 5 months ago

SillyTavern - 1.12.0

Published by Cohee1207 5 months ago

SillyTavern 1.12.0

Important news

This is an incremental update over the 1.12.0-preview and includes several breaking changes.

View the Migration Guide for details on how to prepare for the update.

Also, several new features require having a modern web browser that supports CSS Nesting. If you experience visual glitches - update your browser and/or operating system to the latest available version.

External media in chats is now blocked by default. Enable it manually if required.

Improvements

  • Migrated to a new format for user data storage and added support for user accounts. See the documentation: Users.
  • Implemented the Data Bank: a method for adding external sources of LLM knowledge. See the documentation: Data Bank.
  • Added vLLM as a Text Completion source and Groq as a Chat Completion source
  • Added new models from OpenAI (GPT-4o), PerplexityAI (Llama 3 Sonar), and Google (Gemini Flash 1.5).
  • Added an ability to select OpenRouter providers in both API modes.
  • TabbyAPI: added min_tokens and banned_strings parameters.
  • LlamaCpp: allowed JSON schema usage (conflicts with grammar).
  • Added the AI assistant contest winner character card (Sakana) to the "Download Extensions & Assets" menu and first run onboarding dialog. There you can also view and download past contest winners.
  • World Info improvements: added ability to activate entries using the Vector Storage extension, Example Messages insertion position, weighted and scored inclusion groups, prioritized entries in inclusion groups, support for multiple inclusion groups per entry, ability to delay the entry evaluation until the first recursion step, regular expression type for the entry key, entry duplication, and many more. See the documentation: World Info.
  • Settings are now backed up automatically every 10 minutes instead of on startup. Increased the limit of snapshots to 50 per user.
  • Added Llama 3 model tokenizer. Claude tokenizer is now initialized lazily.
  • Added instruct/context templates for the Phi model, and variants of Llama 3 and ChatML with message names inclusion.
  • The "Advanced search" option now sorts the search results by relevance.
  • Tags & Folders: added the ability to show only or hide folders. Added a toggle for alphabetical tag sorting.
  • Added the ability to hide avatar images in the chat. Reorganized User Settings menu.
  • Added the ability to import from AICharacterCards.com and generic URLs using the external import dialog.
  • Added a quality setting for image inlining in Chat Completion.
  • Forwarded IP whitelisting: can now be disabled with a configuration option.
  • Added a console script for server plugin management.
  • Various localization improvements.

STscript

  • A brand new parser with support for closures, auto-complete, and syntax highlighting. See the documentation: STscript.
  • Added a "negative" argument to the /imagine slash command.
  • Added "role" and "hidden" arguments to the /messages slash command.
  • Added optional "name" argument to the /send slash command.
  • Added the /renamechat slash command.
  • /summarize can now summarize any text, not just the chat messages.
  • /sendas old syntax is now deprecated. Use the new syntax: /sendas name=<character> <message>.

Extensions

  • Image Generation: added new NovelAI model, prompt prefixes now support macros.
  • Vector Storage: added settings for WI activation and Data Bank inclusion.
  • Chat Translation: added an option to translate user input text.
  • TTS: MsEdgeTTS is now supported via the server-side plugin. Added a universal playback speed setting.
  • Regex: added an ability to run Regex on the WI entries.
  • Macros: added {{char_version}}, {{timeDiff}} macros.
  • Added more downloadable extensions to the "Download Extensions & Assets" menu. Check them out!

Bug fixes

  • Fixed InfermaticAI non-streamed response parsing.
  • Fixed start.sh script to allow starting the server from any directory on Mac.
  • Fixed a vertical alignment of zoomed avatars in visual novel mode on phones.
  • Fixed the main prompt being purged from the prompt manager when disabled.
  • Fixed the "Replace/Update card" function with UTF-8 characters in the filter name.
  • Fixed text copy not working on messages hidden from the prompt.
  • Fixed whole-word WI search to include punctuation and other non-alphanumeric characters.
  • Fixed Trigger% being displayed as 0% when imported from external lorebooks.
  • Fixed the "Narrate translated" option not working with TTS.
  • Fixed first message regeneration on character modification if the message was edited.
  • Fixed boolean values matching when searching for WI entries with /findentry.
  • Fixed custom ordering of WI entries being able to sort past the header element.
  • Fixed auto-selection of settings presets by name match (now requires an exact match instead of partial).
  • Fixed multiswipe interaction with response streaming.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.8...1.12.0

SillyTavern - 1.12.0-preview

Published by Cohee1207 6 months ago

SillyTavern 1.12.0 - Neo Server Preview

This is a beta pre-release version. While it's stable enough for everyday use, it still may have some bugs and was not tested on Termux or with ST Launcher. Make sure to report any encountered issues.

To install this version, please switch to the staging branch:

git switch staging

Breaking changes

Data migration

This update changes the way SillyTavern handles user data and introduces a multi-account mode.

/public directory no longer stores user data.

You may have to do some preparation work before running an updated version!

Please refer to the documentation for more information: SillyTavern 1.12.0 Migration Guide

Since the default UI themes, backgrounds, QR and Moving UI presets are removed from /public, you may have git merge conflicts on pulling. Don't forget to back up important data, then run git reset --hard, then try pulling again.

External media

As announced in release 1.11.7, external media in chats is now blocked by default. Please enable it manually per character if needed.

Security notice

NEVER HOST ANY INSTANCES TO THE OPEN INTERNET WITHOUT ENSURING PROPER SECURITY MEASURES FIRST.

While we tried to implement the best security practices, hosting publicly accessible SillyTavern instances is not recommended.

If you still want to set an externally accessible instance, please refer to our guide: Reverse Proxying SillyTavern

New features

User Data & Accounts

  • Introduced a configurable path for all persistent user data. You can now store user data outside of the repository, like AppData or dot directories.
  • Added a local multi-account mode with optional password protection and an admin role. Every user has an independent data directory.
  • Added a two-mode login screen when multiple users are configured: account picker or discreet login.
  • CSRF protection no longer prevents from opening an app in multiple tabs. Use with care, data sync between several open tabs is still not implemented.
  • Added an "Account" dialog under the user settings menu. There you can:
    • Download a full backup of all user data from the UI.
    • Reset just your settings or everything to factory defaults.
    • Manage setting snapshots.
    • Change your password (if enabled).
  • Added an "Admin Panel" dialog under the user settings menu. There you can manage existing or create new users.

Data Bank (RAG)

Data Bank is a new and improved way of using the Vector Storage plugin for Retrieval-Augmented Generation. Build your library of documents and seamlessly integrate them into LLM prompts.

See the documentation for a comprehensive description: Data Bank

  • Added a unified document storage with three scopes: global, character, and chat.
  • Implemented diverse ways of building data collections: local files, web pages, Fandom wikis, YouTube videos, etc.
  • Added Cohere API as the source of embeddings.
  • Added an experimental option to translate documents to English before generating vectors.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.8...1.12.0-preview

SillyTavern - 1.11.8

Published by Cohee1207 6 months ago

SillyTavern 1.11.8

Backends

  • Perplexity: added as a Chat Completion source.
  • OpenAI: added new GPT-4 Turbo models.
  • Google MakerSuite: added support for system prompt usage. Unlocked context size now goes up to 10⁶ tokens.
  • Custom Chat Completion source can optionally use a Claude prompt converter.
  • TabbyAPI: added a setting for JSON schemes.

Improvements

  • Increased default character avatar size from 400x600 to 512x768.
  • Group chats: "Join character cards" mode can now define a prefix and suffix for every merged field.
  • Instruct mode: added templates for Llama 3 Instruct and Command-R. Newline is no longer forced at the end of the story string when newline wrapping is disabled.
  • Prompt manager: chat history and examples can be disabled. Moved the prompt controls bar to the top.
  • Macros: added a {{noop}} that resolves to an empty string. {{trim}} macro can now be used in the chat start field.
  • The majority of token calculations are now asynchronous and won't block the UI when counting.
  • Added an option to enable magnification on zoomed avatars.
  • Whitelisting: added a check for forwarded IPs.
  • Updated the visual layout of character tag controls and API setting preset controls.
  • Various localization fixes and improvements.

Extensions

  • Image Generation: added Pollinations as a source. %char_avatar% and %user_avatar placeholders are now available for the ComfyUI workflow editor (replaced with data URI encodings of respective images).
  • Vector Storage: added experimental setting to summarize messages before embedding.
  • Quick Replies: added ability to set tab size and editor and use a Ctrl+Enter hotkey to execute a script being edited.
  • Character Expressions: added classification using an LLM prompt and a setting for fallback expressions.

STscript

  • Added /caption command for Image Captioning extension.
  • /bg without arguments now reports a current background name.
  • /proxy without arguments now reports a current proxy name.
  • /cut command now outputs the text of removed messages to the pipe.
  • /random command with a tag name provided as an argument can now pick a random character with a specified tag.

Bug fixes

  • Fixed the centering of the load spinner.
  • Fixed Ctrl+1-9 hotkeys being intercepted while not doing anything.
  • Fixed regenerate removing more than one message when a non-streaming API fails to produce any text.
  • Fixed recursive split function (used by various extensions) producing duplicate chunks.
  • Fixed loading of server plugins that provide lifecycle functions in default exports.
  • Fixed behavior of {{lastMessage}} and {{lastMessageId}} macro during swiping.
  • Fixed forced persona name not being added to example dialogues in instruct mode.
  • Fixed {{pick}} macro rerolling on branches and renamed chats.
  • Fixed empty lines produced by the "Join character cards" setting in groups.
  • Fixed macro not being substituted in example separators and story strings.
  • Fixed interaction between TTS and streamed generations.
  • Fixed substitution of macros in TTS text before narration.
  • Fixed highlighting of newly added characters.
  • Fixed doubled token counting of in-chat injections for prompt message fitting.
  • Fixed WI recursion override checkboxes missing in localized versions.
  • Fixed missing names in example dialogues for Cohere prompts.
  • Fixed version display in the welcome message.
  • Fixed performance of /hide and /unhide commands.
  • Fixed image generation with the Draw Things app.
  • Fixed line breaks encoding in message-embedded style tags.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.7...1.11.8

SillyTavern - 1.11.7

Published by Cohee1207 7 months ago

SillyTavern 1.11.7

Important news

  1. This update includes a substantial update to the Instruct Mode formatting. Since all of the default templates were updated, you may experience merge conflicts on git pull if you updated default instructs/contexts. To fix, make a backup, then do git reset --hard before pulling again.
  2. You may also need to adjust your custom instruct templates to the new format if you previously used workarounds for ST's format limitations. See the documentation for more info: https://docs.sillytavern.app/usage/core-concepts/instructmode/
  3. The "Tags as Folders" setting now requires manually enabling which tags are classified as folders. You can do this in the Tag Management menu.
  4. External media files can now be individually allowed/forbidden for every character and group. The setting to forbid external media will be enabled in the next release by default. Make sure to manually enable media in advance where required.

Backends

  • TextGen: added smooth curve support.
  • MakerSuite: added Gemini 1.5 models.
  • Cohere: added as a Chat Completion source.
  • Mancer: enabled logprobs viewer, dynamic temperature, and multiswipe.
  • llamacpp: added optional API key support.
  • TabbyAPI: added an ability to send JSON schemas.

Improvements

  • Reworked Instruct Mode: added separate message wrapping sequences for all message roles, added user filler message, added ability to restore default templates.
  • Added per-character and per-group override controls over the in-chat external media removal. Media: images, audio, video. External: not hosted by the local server.
  • Improved visual display and usability of Tag Folders with hidden folders and inline characters display. Added three-state mode for all tag filters and tag sorting.
  • Added an ability to expand left-side extensions windows (Author's Note, CFG, Logprobs).
  • Added pick macro: an alternative to random with the same syntax that preserves its selection on consecutive evaluations if the source string is unchanged.
  • Added experimental Smooth Streaming mode. Text is streamed letter by letter and stops on punctuation. Logprobs view is disabled in this mode.
  • Added ability to select a role for at-depth chat injection (system/assistant/user) for World Info, Author's Note, Character Note, Inject command, and Summary.
  • Allow AVIF and WEBP formats for user and character avatar images. WEBP character card import is still unsupported.
  • Chat Completion: added control of continuation append postfix - space (old behavior), newline, or double newline.
  • Prompt Manager: added an indication of prompts overridden by a character card; added ability to prepend names to message contents; added ability to disable the Main prompt; a selected custom prompt dropdown option is now preserved on render;
  • Auto-continue: now allowed in group chats.
  • The server port and listen mode can now be controlled with console arguments when starting up.
  • Added ability to delete UI themes.
  • Added lastCharMessage and lastUserMessage macros.
  • Various localization fixes and improvements.

Extensions

  • Duplicate Finder: added to the downloadable extensions list.
  • Assets: added filters to the list of available assets.
  • Quick Reply: added word wrap toggle for QR editor.
  • Image Generation: added Draw Things as a generation source; SMEA/DYN controls for NAI Diffusion.
  • Summarization: added a raw prompt building mode and auto-adjustment of summarization threshold.
  • Caption: added new multimodal models from OpenRouter.
  • TTS: added a setting to skip tag narration.
  • All messages in the DOM now have a "swipeid" attribute that indicates the currently selected swipe index.

STscript

  • Scripts now support comments: // and /#.
  • Message-sending commands (/send, /sys, etc) now support an optional compact argument for slim in-chat display.
  • /genraw: added "system" argument to control the system prompt.
  • /instruct, /context, and /model commands now return the selected entry name if no argument is provided.
  • Added /classify command to get the emotion label from any text.

Bug fixes

  • Fixed npm packages audit messages.
  • Fixed app startup on Safari when the browser process was killed.
  • Fixed inability to use more than 9 capturing groups in regex.
  • Fixed card tags import when importing multiple cards at once.
  • Fixed auto-scroll interruption not working in Firefox.
  • Fixed NovelAI image generation endpoint.
  • Fixed grammar string not being properly sent to KoboldCpp.
  • Fixed prompt cache not being reset on API error.
  • Fixed double append of character names in example dialogues for Claude when using "add character names" settings.
  • Fixed the first message not updating in pristine chats when using the /persona command.
  • Fixed random macro not working when using both syntax flavors in one string.
  • Fixed UI lag when updating prompt manager prompts.
  • Fixed data sanitation in the Token Counter plugin.
  • Fixed app crash on persona names that contain a single quote in their name.
  • Fixed performance of hotkey and window resize event handlers.
  • Fixed auto chat load not considering characters/groups opened with /go and /random commands.
  • Fixed first messages not getting auto-translated when creating a new group chat.
  • Fixed tags not being transferred to duplicated characters.
  • Fixed squashing of system messages adjacent to example dialogues.
  • Fixed "Raw last message" image generation mode not working in groups.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.6...1.11.7

SillyTavern - 1.11.6

Published by Cohee1207 7 months ago

SillyTavern 1.11.6

Important: if you're having issues with installing NPM packages after updating, make sure you're using the latest LTS version of Node (v20.11.1 as of writing).

Improvements

Generation APIs

  • Breaking: switched Claude to Messages API. You may need to adjust your prompts.
  • Claude: added Haiku, Sonnet and Opus models, including Vision capabilities and multimodal captioning.
  • TextGen WebUI: added max_tokens_second control.
  • Aphrodite: added text generation model selector.
  • DreamGen: added as a Text Completion source.
  • KoboldCpp: added API key/password support, added as a multimodal captioning source.
  • Chat Completion: updated default sampling parameters to more neutral (1 temperature, 0 penalties).
  • OpenRouter (Text Completion): added maximum prompt cost calculation.
  • OpenRouter (Chat Completion): added deprecation notice for instruct override mode.
  • Request overrides: can now apply additional headers to all Text Completion sources.

General UX and usability

  • Added Portuguese, Vietnamese, Ukrainian, French, Icelandic and Arabic translations.
  • Added ability to use __underlined__ text formatting in messages.
  • Added ability to view Creator's note from the character panel.
  • Added ability to replace the character card with a selected file rather than importing it as a copy. Opens as "Replace / Update" from the "More" menu.
  • Context formatting: added a toggle to use jailbreak from the character card with non-Chat Completion APIs.
  • Added lazy evaluation of macros used in character's chat greetings. This allows setting chat variable values that are unique to the chosen greeting option.
  • Allowed to use @import in custom CSS of themes. Importing themes that use imports requires user confirmation.
  • Instruct mode: added an option to disable example dialogue formatting. Added special {{name}} macro for instruct sequences.
  • Added ability to open the character source URL from the "More" dropdown menu (only Chub and Pygmalion imports).
  • Removed restriction for the used model for multimodal inlining with Custom Chat Completion endpoint.
  • World Info recursed entries now evaluate macro before being scanned.
  • Added ability to import JSON chats from Agnai.
  • Now using Readability to extract text from HTML pages.
  • Official startup scripts now run Node in the production environment.
  • Changed the icon for the "Save as new preset" buttons to a more obvious one.
  • Reduced motion toggle now also disables chat background transitions.
  • Disabled character import and creation during the message generation to prevent data corruption.

Extensions

  • General: added an API for extension authors to write character card additional fields.
  • Token Counter: added debouncing of tokenization requests.
  • Vector Storage: added TogetherAI and NomicAI as embedding providers, added new OpenAI models selection.

STScript

  • Added /model command: select the model for the current API.
  • Fixed /fuzzy command: improved accuracy of search.

Bug fixes

  • Fixed not being able to use the max value for WI entry scan depth.
  • Fixed group avatars when no members are left in the group.
  • Fixed uploading images if the character name contains a dot.
  • Fixed Scale generation using JWT tokens.
  • Fixed link to a Horde worker repository.
  • Fixed unlocked context size uncheck when loading a preset without max context saved.
  • Fixed background images rarely not filling the whole page when the app is loaded.
  • Fixed request overrides headers not being applied on tokenization requests.
  • Fixed parsing of STscripts containing spaces in named arguments.
  • Fixed STscripts commands accepting numeric ranges erroring when math commands output is being passed.
  • Improved performance of timestamp parsing.
  • Improved performance of macro evaluation.
  • Improved performance of loading very large untruncated chats.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.5...1.11.6

SillyTavern - 1.11.5

Published by Cohee1207 8 months ago

SillyTavern 1.11.5

Improvements

  1. Added OpenRouter as a Text Completion source to benefit from more precise Instruct formatting.
  2. Regex scripts and UI themes can now be imported/exported via JSON.
  3. Added sampler priority for Text Generation WebUI and llama.cpp backends.
  4. SillyTavern repository is now excluded from Android gallery media scans.
  5. {{group}} macro can now be used anywhere to get a comma-separated list of group member names.
  6. Added an option to preserve the pixel count when using image generation modes that change the aspect ratio.
  7. -1 is now a min value for Top K slider in Text Completion (behavior differs across backends).
  8. Added InfermaticAI as a Text Completion source.
  9. Added import of characters from Pygmalion.chat by URL.
  10. Added support for character card type assets in asset downloader. Any new default AI characters (if any) will be distributed with it.
  11. Bumped unlocked context size for Chat Completions to 200k.
  12. /impersonate and /continue commands now accept an optional 'quiet' prompt to use with respective generation modes.
  13. Slash commands are no longer executed from the chat bar inputs if the slash is not the first entered character.
  14. Per-entry scan depth override max value is now 1000. Added warnings if the value is overflown by user input.
  15. Translation provider URLs can now be reviewed in the UI once saved.
  16. Added logprobs support for llama.cpp, Aphrodite and Tabby.
  17. Added ability to execute STscripts when World Info entry is activated using the Automation ID field input.
  18. Added input autocomplete for WI Inclusion Group and Automation ID.
  19. Added mistral-large model to the MistralAI selection list.

Bug fixes

  1. Fixed regex application for converted group chats.
  2. Fixed squashing system messages if there are any empty system messages in the completion.
  3. Fixed renaming of the current chat file.
  4. Fixed single quotes being removed from generated image prompts.
  5. Fixed AllTalk TTS connection to remote servers.
  6. Fixed status check firing for Chat Completion on load even if another API is currently selected.
  7. Fixed sorting by "Newest" if the V2 spec character card was imported between May to September 2023 and was never interacted with.
  8. Fixed macros substitution in the "New Example Chat" utility prompt and edited messages.
  9. Fixed Mistral and Extras vectors not getting purged.
  10. Fixed vector insert count settings being unused (previously used query count for it).
  11. Fixed keep_alive parameter not being sent out to Ollama backend.

What's Changed

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.4...1.11.5

SillyTavern - Docker fix test

Published by Cohee1207 8 months ago

Do not use.

SillyTavern - 1.11.4

Published by Cohee1207 8 months ago

SillyTavern 1.11.4

  1. Added API key support for Text Generation WebUI.
  2. Added /proxy command.
  3. Added await argument to /trigger command.
  4. Added /getchatname command.
  5. Fixed /delchat command performance.
  6. Added textgen logprobs to TabbyAPI.
  7. Switched MakerSuite streaming to SSE.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.3...1.11.4

SillyTavern - 1.11.3

Published by Cohee1207 9 months ago

SillyTavern 1.11.3

Improvements

  1. New and improved UX for the Persona Management panel.
  2. Added per-entry setting overrides for World Info entries.
  3. Scan Depth in World Info now considers individual messages, not pairs.
  4. Added logprobs display for supported APIs (OpenAI, NovelAI, TextGen).
  5. Added repetition penalty control for OpenRouter.
  6. Added sanitation of external media in chat messages (optional for now).
  7. Added presets for reverse proxies and MistralAI proxy type.
  8. Added new OpenAI models to the list.
  9. Allowed to use multiple stop strings for TogetherAI.
  10. Improved UI tooltips for advanced settings.
  11. Added quad sampling controls for supported Text Completion sources.
  12. Renamed Roleplay instruct template to Alpaca-Roleplay.
  13. Removed the old format of setting presets from the repository.
  14. Improved behavior of sentence trimming. Punctuation is now trimmed if preceded by a whitespace.
  15. Aphrodite: added tokenization, grammar, and dynamic temperature.
  16. OpenAI: added ability to generate multiple swipes per request.
  17. Groups: added join cards prompt mode that keeps the disabled members.
  18. Reverse proxy controls are moved to the API connection panel.

Extensions

  1. Added VRM extension to the registry.
  2. TTS: added AllTalk (external server) and SpeechT5 (built-in) TTS providers support.
  3. Vector Storage: added bulk embedding calculation, added Extras API as vectors source.
  4. Speech Recognition: added built-in Whisper recognition provider (no Extras required).
  5. Regex: added min/max depth for prompt and display regex scripts. Removed Overlay mode (it did nothing since 1.11.2).
  6. Chat Translation: added Lingva translation provider support.
  7. Image Generation: /sd slash command now returns a link to the generated image and accepts the quiet=false argument to suppress chat images.
  8. Timelines: various UX improvements.
  9. Character Expressions: added refresh of sprites on upload (only Chromium).

STscript

  1. Every command can now accept backslash-escaped macro and pipe-separators in named and unnamed arguments.
  2. Added /instruct and /context commands that set the appropriate templates.
  3. Added ability to execute QR on group member trigger.
  4. Added /chat-manager command.

Bug fixes

  1. Fixed opening chats when using search in chat manager.
  2. Fixed formula rendering displaying the contents twice.
  3. Fixed scroll bars display in Chrome 121.
  4. Fixed interaction with double quotes inside of HTML/XML tags in messages.
  5. Fixed non-unique chunks being inserted by Vector Storage.
  6. Fixed console nag due to tags v2 fields mismatch.
  7. Fixed installed extensions persistence in Docker.
  8. Fixed numeric zeros usage in {{pipe}} macro.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.2...1.11.3

SillyTavern - 1.11.2

Published by Cohee1207 9 months ago

SillyTavern 1.11.2

Known issues

  1. "Error rendering template" error appearing in Chrome. This relates to the antivirus (McAfee) browser extension, disable the browser extension if you face this issue.
  2. Overlay mode for Regex scripts currently doesn't work. You can use explicit capturing group references instead (see below).

Improvements

  1. Chats with only one message won't be recreated on making edits to the character card if any generative action is performed.
  2. Multiple extensions can now be disabled before reloading the app.
  3. World Info: Added AND ALL logic for secondary key matching.
  4. World Info: Added "prevent recursion" option for entries.
  5. World Info: Added a button to duplicate a WI file.
  6. Claude: added option to exclude Human/Assistant prefixes from chat history in the prompt.
  7. Text Completion: added Dynamic Temperature controls for supported backends.
  8. Text Completion: API tokenizer can be used to compute logit bias and banned tokens.
  9. Text Completion: added ability to bypass status checks and use a type-in model.
  10. Auto-connect and auto-load no longer use local storage settings.
  11. Instruct-formatted background prompts now use an output sequence with a "System" pseudonym.
  12. Advanced Formatting: added {{mesExamplesRaw}} macro for unformatted example dialogues.
  13. OpenRouter: redesigned model selector, added Min P and Top A sampling parameters.

Extensions

  1. Quick Replies: completely rewritten with the new editor, import/export, multiple active sets, etc.
  2. Captioning: added Text Generation WebUI as a multimodal source.
  3. Web Search: added search querying mode using backticks in user messages.
  4. Vector Storage: added chunking of stored messages, and changed default embedding model.
  5. Regex: fixed matching positions, added capturing group replacements using $1-syntax.
  6. TTS: added chunking for NovelAI source, streaming support, and advanced options for XTTS.

STscript

  1. Added /lastsprite command.
  2. Added ability to navigate to groups with /go.
  3. Added optional arguments to the /world command.
  4. Added argument for /gen to override instruct "System" pseudonym.

Bug fixes

  1. Fixed embedded lorebooks being removed from the card if the linked WI is not locally found.
  2. Fixed ComfyUI image generation.
  3. Fixed server crash when using GUI preset for KoboldAI Classic.
  4. Fixed TTS voice mapping for groups.
  5. Fixed numeric inputs for some WI slider inputs.
  6. Fixed WI inclusion groups behavior with recursion.
  7. Fixed Repetition Penalty being improperly sent to KoboldCpp under Text Completion.

Community updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.1...1.11.2

SillyTavern - 1.11.1

Published by Cohee1207 10 months ago

SillyTavern 1.11.1

This is the last stable release of 2023. See you in 2024!

API sources

  1. Added Google MakerSuite API (Gemini models), replacing PaLM API.
  2. Added "Custom" Chat Completion source for "OpenAI-compatible" endpoints, intended to replace OpenAI with a reverse proxy (this is still an option). It's impossible to provide compatibility with everything. We can't guarantee that your specific endpoint will work.
  3. Added new Text Completion sources:
  • MistralAI
  • TogetherAI
  • Ollama
  • llama.cpp
  1. Added server-side plugin API for extension authors. See the docs: https://docs.sillytavern.app/for-contributors/server-plugins

Improvements

  1. Added a toggle for single-row input form buttons display on phones.
  2. Added a toggle for reduced motion that disables most UI transitions.
  3. Added an alternative continue prompt method for Chat Completions.
  4. Added FPS limit for text streaming (default: 30 t/s).
  5. Added an option to skip TTS narration of code blocks.
  6. Increased height of advanced definitions rows.
  7. Configurable auto-mode delay for group chats.
  8. Added confirmations for convert to group and hotkey regeneration.
  9. Claude: improved system prompt usage. Added validation of prompts.
  10. Context and instruct presets are now decoupled by default. The old behavior is optional.
  11. Added support for TextGen WebUI yaml characters import.

Extensions

  1. Vector Storage:
  • recalled messages can now activate World Info entries.
  • added MistralAI source.
  1. Web Search: added "visit mode" to download the content of found web results.
  2. Token Counter: added token ids display for KoboldCpp.
  3. Summarization: simplified settings for Extras source, buffer size is now always ~1024 tokens.
  4. Caption:
  • added new multimodal sources (Ollama, llama.cpp, Google MakerSuite, OpenAI proxy).
  • added an option to always ask for the prompt before multimodal captioning.
  1. Image Generation:
  • added TogetherAI source.
  • added character-specific negative prompts.
  • added 'char' prefix / {{charPrefix}} macro to free mode.

STscript

  1. Added /preset command to switch the settings preset for an active API.
  2. Added a set of commands for managing Quick Replies.
  3. Added /rand command that returns a random number.
  4. Added the /icw command to switch ComfyUI workflows.

Bug fixes

  1. Random sorting of characters list should now be more random.
  2. Persona name now properly updates when using an old-style user name input.
  3. Fixed display of large images inserted with Markdown.
  4. Fixed continue on user and system narrator messages in group chats.
  5. Fixed UI hangup on TTS engine initialization.
  6. Fixed performance of stop strings clean-up during streaming.
  7. Fixed generation abortion for KoboldCpp via Text Completions.
  8. Fixed request body size limit for CORS bypass endpoint.
  9. Improved precision of max context slider.
  10. TXT export doesn't include invisible system messages anymore.
  11. Fixed source-specific controls display for Summarization extension.
  12. Fixed offscreen group member management with slash commands.
  13. Fixed Llama2Chat instruct preset
  14. Fixed style element sanitation for chat rendering.
  15. Fixed duplication of custom expressions.
  16. Added chunked translation of long texts via the extension.

Community Updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.11.0...1.11.1

SillyTavern - 1.11.0

Published by Cohee1207 11 months ago

SillyTavern 1.11.0

Important news

  1. This update includes a security patch. Don't ignore it if you're hosting publicly accessible (e.g. tunneling or port forwarding) SillyTavern instances. What's changed: removed the ability to issue CSRF tokens without passing authentication or whitelist check first. This can only create a minor inconvenience when a legitimate user needs to refresh a page to reissue the token. Privacy of the user data is not at risk.
  2. The majority of the internal API routes were changed. If you are a third-party extensions developer and use an internal HTTP API in your plugins, consider migrating to new endpoints. Temporary redirects will eventually be removed.
  3. API setting preset file type was changed from .settings to .json and copied to the /default/ directory within the repository. Custom presets will be automatically converted to the new format next time you run SillyTavern. Old format presets will be removed from the repository in the next release and you will have merge conflicts on pull if you modified default presets. To avoid this and revert changes to old default presets, backup them and run git reset --hard.

New features

  1. Added character import from JanitorAI by URL via the "Import external content" menu.
  2. Added pop-out menu for Quick Replies. QR editor can be expanded and have custom tooltips.
  3. Added ability to customize previously hidden Prompt Manager bits: scenario wrapper, personality wrapper, and group chat nudge.
  4. Added restoration of unsent user inputs when reloading the page. It can be disabled via "User Settings".
  5. Increased the "Response (length)" slider max value to 2k by default and 16k when using the unlocked context option.
  6. Added sampler seed control for OpenAI API.
  7. Added ability to enable user ID randomization for OpenAI API via config.yaml.
  8. Added "Universal" presets for KoboldAI and Text Completion that use Min P sampler.
  9. Added the ability to use HTML, PDF, and Markdown files with the "Attach file" function.
  10. Added chunked RAG for attached files to the "Vector Storage" plugin.
  11. Added Min P sampler to Simple UI mode.
  12. Web Search extension can use Extras API with Selenium browser (also available on Extras Colab).
  13. Added context separator strings to a list of stop strings. It can be disabled via "Advanced Formatting".
  14. Added "formality" setting control for DeepL translation. Configurable via config.yaml.
  15. Always prefer API tokenizer in "Best match" mode if supported.
  16. Added ability to open an expanded editor of "Description" and "First Message" fields.
  17. Renamed "Bookmarks" to "Checkpoints".
  18. Added ability to restore default setting presets.
  19. Added multiple response generation as swipes for Aphrodite API source.
  20. Added inclusion groups for World Info entries.
  21. Added NOT ANY logic for the World Info secondary filter.

Slash commands / STscript

  1. Added a set of arithmetic commands (/add, /div, /mod, etc).
  2. Added "at" argument to commands that add a chat message to a designated place.
  3. Added /inject command for custom prompt injections.
  4. Added commands and macros to increment and decrement variable values.
  5. Added /regex command to run Regex scripts by name.
  6. Now /trigger command can be used outside of group chats.
  7. Added a set of commands to read/write World Info files.
  8. Added {{maxPrompt}} macro to get a max allowed prompt length in tokens.
  9. Added /forcesave command to save the chat and settings.
  10. Added a set of commands for JSON-serialized array and object variables.
  11. Added customization arguments to user I/O commands.
  12. Added {{currentSwipeId}} / {{lastSwipeId}} macros.
  13. Added ability to use /run with any QR preset using /run setName.scriptName syntax

Please see the docs on how to use STscript: https://docs.sillytavern.app/usage/st-script/

Bug fixes

  1. Fixed Silero TTS generation when using a standalone server.
  2. Fixed the inability to select menu items in the wand menu with long taps.
  3. Fixed character sprite appearing in non-VN mode behind the chat block in the mobile layout.
  4. Fixed using CORS bypass middleware with the request body.
  5. Fixed XTTS text preprocessing which previously caused unnatural sounding noises.
  6. Fixed recursion in QR auto-execution if the STscript command sends another message.
  7. Adjusted max context limit for PaLM API due to using GPT tokenizer.
  8. Fixed {{mesExamples}} macro not working in story string with "Strip examples" behavior.
  9. Fixed display of zoomed avatars on ultrawide screens.
  10. Fixed removing trailing stop strings when cleaning up the output text.
  11. Require the generation to be unlocked to run /trigger or /continue commands.
  12. Fixed NOT ALL World Info filtering logic.
  13. STscript auto-execute commands are now run after the message is rendered, not before.
  14. Fixed {{original}} macro not being removed if the prompt is empty.
  15. Reverted order of {{random}} macro evaluation to pre-1.10.10.
  16. Fixed performance of bulk favorite and bulk delete.

Community Updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.10.10...1.11.0

SillyTavern - 1.10.10

Published by Cohee1207 11 months ago

SillyTavern 1.10.10

NB: config.conf has been migrated to config.yaml format. It should auto-convert the next time you run npm install.

New Features

  1. Slash commands are now a part of STscript - an embedded scripting language for almost unlimited customization. See the guide to get started: https://docs.sillytavern.app/usage/st-script/
  2. Added ability to attach text files (sent in prompt, verbatim) and images (for image inlining) directly to messages, omitting captioning.
  3. Prompt Manager - all non-essential prompt bits can now be disabled individually.
  4. World Info - added exclusion/filtering by character tags.
  5. Added backup/restore for tags and personas.
  6. Renamed TextGeneration WebUI to Text Completion API. KoboldAI was renamed to KoboldAI Classic and is now feature-frozen.
  7. Added KoboldCpp and TabbyAPI sources to Text Completion API.
  8. Added Yi model tokenizer.
  9. Added configurable chat truncation amount for large chats, previously hard-capped at 100.
  10. Added more information to the Mancer model selection dropdown.
  11. Added Claude v2.1 and system prompt formatting.
  12. Moved wand menu to the left, optimized mobile chat bar layout, and added square avatar style.

Extensions

  1. Web Search - new extension in the downloadable extensions index. Allows to add organic Google search results to LLM prompts. Requires a SerpApi key.
  2. Image Captioning - decoupled from image inlining, added OpenRouter as a multimodal source (gpt-4 and llava-13b models).
  3. EmulatorJS - added an interactive comments mode: characters will provide witty comments based on snapshots of your gameplay. Requires a multimodal captioning or image inlining enabled (OpenAI and OpenRouter are supported).
  4. Speech Recognition - added OpenAI Whisper API (requires an API key).
  5. Quick Replies - added auto-execution on set conditions (useful for scripts) and nested modal menus, QR slots are now drag sortable, added button to update current QR preset.
  6. Image Generation - added ComfyUI source, added NAI Diffusion v3 model, and added templates for common image resolutions.
  7. Added CORS bypass endpoint (required for some 3rd-party extensions, enabled with config.yaml).
  8. TTS - added XTTSv2 server as a source.

Bug fixes

  1. Fixed chat deletion when no messages are prompted for deletion.
  2. Fixed file extension of chat file backups (.json => .jsonl).
  3. Fixed World Info NOT operator for secondary keys.
  4. Fixed low-res avatars on Image Generation results.
  5. Fixed server crash when uploading large images.
  6. Fixed character name appending when continuing user message.
  7. Fixed space trimming on character name clean-up.
  8. Fixed PaLM API payload validation error.
  9. Fixed server crash on Google Translate error.

Scripting / Slash Commands

  1. Added flow control commands.
  2. Added chat variables command and macros.
  3. Added text manipulation commands.
  4. Added user input/output commands.
  5. Added commands to manipulate the group members list.
  6. Added commands to remove/delete swipes.
  7. Added macros for instruct sequences.
  8. Added alternative syntax for random macro.

Community Updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.10.9...1.10.10

SillyTavern - 1.10.9

Published by Cohee1207 11 months ago

SillyTavern 1.10.9

Bug Fixes

  • Fixed setting sliders breaking in non-English locales.
  • MinP sampler for Kobold is now hidden behind a feature flag.
  • Fixed interactive mode for SD not triggering with certain phrases.
  • Fixed debouncing duration of number inputs.
  • Fixed unlocked context size counter.
  • Added null checks for OpenRouter models pricing.
  • Added blocking loader for mass deletion to prevent data corruption.

What's New

  • Added image inlining for the GPT4-V model (sending prompts with pictures).
  • Added selector for auxiliary list view filed (author or version).
  • Returned ability to customize WI entry order ("Custom" sorting).
  • Added OpenAI TTS provider.
  • Added ability to bypass OpenAI status check.
  • Added prompt to create a persona when uploading an avatar.
  • Added grouping and sorting to OpenRouter model selector.
  • Live2D models can be grouped into folders. This is a breaking change. Please update the Live2D extension via the Manage Extensions menu.

Community Updates

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.10.8...1.10.9

SillyTavern - 1.10.8

Published by Cohee1207 11 months ago

SillyTavern 1.10.8

New Features

APIs:

  • Added new OpenAI models (GPT3.5/GPT4 Turbo, DALL-E - see below).
  • Added forced instruct toggle for OpenRouter to use instruct mode formatting.
  • Added Mistral model tokenizer. OpenRouter models will use a correct tokenizer if available.
  • Updated TextGeneration WebUI support to use a new API schema. Legacy API can be used with a toggle (without streaming).
  • Rearranged API setting inputs for Kobold and TextGen for a more compact display with on-hover help, and added Min P sampler.
  • Added Zen Sliders (compact mode) and Mad Labs (unrestricted mode) for Kobold and TextGen settings.
  • Refactored status checks, and added an ability to cancel a pending API connection.
  • Added check for filtered Google PaLM API response.
  • Added separate API icons for Aphrodite and Mancer.

Slash commands:

  • Can now be run with quick replies while the generation is in progress.
  • Added /trigger slash command to trigger a group member reply (by order or by name).
  • Added autocomplete for slash commands by typing / in chat.
  • Added hide/unhide slash commands and a message action button to hide the message from the prompt (indicated by the ghost icon).
  • Added /speak slash command to trigger a TTS generation.
  • Added message ranges to /cut command.

Extensions:

  • [Token Counter] Added Token IDs display and Token Breakdown view (if supported by the tokenizer).
  • [Stable Diffusion] Renamed to "Image Generation".
  • [Image Generation] Added DALL-E 2/3 support from OpenAI (API key required).
  • [Image Generation] Added interactive mode to trigger images by sending a chat message - see docs.
  • [Image Captioning] Added support for the GPT-4-Vision model (with customizable prompt).
  • [Image Captioning] Added ability to change chat injection template.
  • [RegEx] Added ability to quickly toggle scripts on and off.
  • [RegEx] Added regex mode to run only on chat messages in prompts.

QoL:

  • Added a display mode to show tags as virtual folders in the characters list.
  • Updated bulk edit mode to mass tag, delete, and favorite characters.
  • Added macros to insert parts of the character card (description, scenario, etc) - see docs.
  • Added a dynamic "Min Activations" scanning range extension for World Info checks.
  • Added lastMessageId macro to insert an ID of the last in-chat message.
  • Extended input length limits for most text boxes.
  • Added page-wide loader to block the UI until the app is fully loaded.
  • Overhauled past chats view and renamed the menu item to "Manage chat files".
  • Overhauled extensions panel.
  • Itemized prompts are now saved between sessions.
  • Creating a dummy persona now requires entering a name.
  • When editing custom prompt manager injection, hide depth input block for relative position.

Bug fixes

  • Removed unreliable check for git installed in the system.
  • Fixed expressions getting stuck sometimes when switching chats.
  • Fixed double-emit of CHAT_CHANGED on /go command
  • Fixed BG top bar icon being on top of moving UI blocks.
  • Fixed WI editor breaking if the entry is not found in the WI file.
  • Fixed macros replacement in the first message if there are alternate greetings.
  • Fixed transparency for zoomed avatars.
  • Fixed /del command deleting all messages on counter overflow.
  • Fixed sending empty bad_word_ids lists to NovelAI.

Community Updates

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.10.7...1.10.8

SillyTavern - 1.10.7

Published by Cohee1207 12 months ago

SillyTavern 1.10.7

This is a minor hotfix/maintenance release.

  1. Addressed common complaints about the behavior of the context length slider:
    • fixed min context value 0 => 512,
    • removed 8k min limit for unlocked context,
    • adjusted slider steps for both unlock states,
    • increased debounce duration for type-in values 500ms => 2sec.
  2. Fixed rare server crash on writing to stats.json file.
  3. Fixed numeric display next to World Info config sliders.
  4. Added sampler seed control for KoboldAI API.
  5. Added a notice that downloading extensions requires to have Git installed.

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.10.6...1.10.7

SillyTavern - 1.10.6

Published by Cohee1207 12 months ago

SillyTavern 1.10.6

Highlights

Most of the previously built-in extensions have been converted to downloadable add-ons. You can download them via the built-in "Download Extensions and Assets" menu in the extensions panel (stacked blocks icon in the top bar).

This includes:

  1. Dynamic Audio
  2. HypeBot
  3. Idle
  4. Speech Recognition
  5. Chat Variables
  6. Parameter Randomizer
  7. Smart Context / ChromaDB
  8. RVC
  9. Objective
  10. D&D Dice

These plugins have been merged with the main code:

  1. Chat Backgrounds (see below)
  2. CFG Scale
  3. Bulk Edit
  4. Settings Search

Apart from old fan favorites, some brand-new extensions are waiting for you there:

  1. Blip - Adds per-word message animation and generated sounds.
  2. Live2D - Enjoy animated models of your characters.
  3. Timelines - New way to interact with chat branching.
  4. EmulatorJS - Play retro console games in your browser.

New features

  1. New generate mode for group chats: "Join character cards" to send all definitions of active members at once. The original mode is now known as the default "Swap character cards" option.
  2. Stable Diffusion plugin improvements:
    • Generated chat backgrounds are now saved to the Backgrounds menu and could be optionally copied to the system backgrounds list.
    • Added NAI Diffusion v2 model for NovelAI source.
    • Added an ability to expand generated prompts via the built-in text generation model based on GPT-2, this technique is similar to Fooocus SD frontend. This works best for SDXL models, and it is recommended to review the prompts before generating when using that feature.
    • Added sanitation of prompts when using Stable Horde to help prevent request flagging. It is highly recommended to leave this on when using SD via Horde.
    • Added the "Style" selector. They act as presets for negative prompts and prompt prefixes.
  3. Chat Completions' prompt manager improvements:
    • Added absolute positioning for custom in-chat prompt injections.
    • Added an option to collapse multiple consecutive system messages into one.
  4. World Info / Author's Note improvements
    • Added ability to include Author's Note into the World Info scanning.
    • Added an option to exclude A/N and WI from the prompt when generating summaries via the main API.
    • Added chat-bound lorebooks.
    • Added option to open the lorebook assignment menu by shift-clicking the globe icon.
    • Added generated prompt display for enlarged chat images.
  5. Added {{// comment}} macro by @IkariDevGIT
  6. Generation API changes:
    • Added stopping strings support for Horde.
    • Added frequency and presence penalty for oobabooga's TextGen.
  7. Added automatic backup of chat files to the /backups/ folder. Can be disabled via the config.conf variable.
  8. Some advanced formatting options (custom stopping strings, newlines handling, trim sentences) can now be saved to a context preset.
  9. Added optional daily notifications about the updates to installed third-party extensions.
  10. The horizontal line element now uses a main text color instead of the border color.
  11. Extended default context size maximum to 8k tokens. Improved slider behavior for AI generation settings.
  12. Added /ask slash command for a one-time query of another character in the context of the current chat by @bdashore3
  13. Dialogue examples behavior is now a tri-state dropdown selector in the power user section (previously in advanced formatting).
  14. Added token counting slash command /count by @city-unit
  15. Added auto-closing of message context buttons when clicking away (use the "Expand Message Actions" option to override).
  16. Added option to use {{input}} macro in quick replies by @IkariDevGIT
  17. New README banner from @deffcolony

Bug fixes

  1. Fixed server crash on providing an invalid streaming Text Generation URL.
  2. Fixed prompt itemization raw text viewer.
  3. Fixed pagination in group members pop-out.
  4. Fixed edge cases of newlines handling for instruct mode.
  5. Fixed token counting for character editor - now counts after replacing macro.
  6. Fixed multilanguage model for ElevenLabs TTS.
  7. Fixed switching between namesake personas.
  8. Improved the correctness of expression classification by increasing the sampling window.
SillyTavern - 1.10.5

Published by Cohee1207 about 1 year ago

SillyTavern 1.10.5

New Features

  1. Added Google PaLM API for Chat Completions and Vectorization.
  2. Redesigned World/Lorebook editor:
    • added at-depth insertion strategy,
    • added sorting order selector,
    • added tri-state entry status,
    • improved embedded lorebook import dialog,
    • added character-specific locks, etc.
  3. Added at-depth Character Notes as an exportable character card field.
  4. Added custom token bans (supports macro command) and grammar support for oobabooga's TextGen.
  5. Split Text Generation WebUI sources (mancer, oobabooga, Aphrodite).
  6. Added new context and instruct templates: Story/Adventure mode, ChatML, Mistral, etc.
  7. Added Bing translation API.
  8. Added SD.Next (vladmandic) Stable Diffusion source.

Bug fixes

  1. Fixed group nudge in Chat Completions.
  2. Increased Stable Horde prompt limits.
  3. Fixed group character expressions and group creation avatar upload.
  4. Fixed bolded text in quotes.

Community contributions

New Contributors

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.10.4...1.10.5

Package Rankings
Top 22.77% on Npmjs.org
Related Projects
UniLLM

🦄 Consume any LLM from any provider, using the OpenAI API

10 Oct 2023 24
casibase

Spising: ⚡️Open-source AI LangChain-like RAG (Retrieval-Augmented Generation) knowledge database ...

29 May 2020 2,751
chat-wechat

06 Aug 2023 11