github SillyTavern/SillyTavern 1.12.5
on GitHub

latest releases: 1.12.7, 1.12.6
2 months ago

SillyTavern 1.12.5

Breaking changes

This release includes several breaking changes to STscript. Please follow the STscript section below to update your scripts.

Backends

  • OpenAI: added chatgpt-4o-latest to the list of models.
  • Block Entropy: added as a Chat Completion source.
  • TogetherAI: now supports min_p, presence_penalty and frequency_penalty sampling parameters.
  • Ollama: now supports min_p sampling parameter. keep_alive setting can be configured in config.yaml.
  • Claude: added a config.yaml option to enable experimental system prompt caching.
  • Perplexity: added Llama 3.1 models to the list.
  • Google MakerSuite: renamed to Google AI Studio.
  • Featherless: model list now shows the maximum context size and automatically sets the context size for the selected model.
  • Added a tokenizer for Gemma and Gemini models.

Improvements

  • Macros in first messages are now synchronized between the display and the prompt.
  • Server: the host protocol can now be configured in config.yaml. Choose between IPv4 (default), IPv6, or both.
  • World Info: disabling an entry is now possible without changing its constant status.
  • World Info: added an option to limit max recursion depth. Conflicts with min activations.
  • Text Completion: Mirostat settings block is now not displayed on sources where it is not supported.
  • Text Completion: names as stopping strings can now be toggled in Advanced Formatting under Context Settings.
  • Chat Completion: group nudge won't be inserted when impersonating in groups.
  • Tags: backup imports can now either overwrite or skip existing tags.
  • Tags: added import options to bulk tag editing menu.
  • Added a configuration option to auto-select the input field contents in popups.
  • Added a configuration option to enable quick impersonation button.
  • Model cache for transformers.js has been moved under the data root directory.

Extensions

  • Image Captioning: raised or removed (if applicable) a token limit for multimodal captioning.

  • Character Gallery: images are now sorted by ascending date.

  • TTS: added an "OpenAI-compatible" endpoint source.

  • Summarization: added an experimental WebLLM extension summarization source.

  • Character Expressions: "LLM" classification source now uses raw prompt generation method.

  • Image Generation:

    • Added Block Entropy and Hugging Face Inference API as image generation sources.
    • Added image swipes for image rerolls and cycling through generated images.
    • Added a separate stop button to cancel image generation (located in the wand menu).

  • Quick Reply:

    • QR panel popup button can now be hidden.
    • Quick Reply buttons can now use an icon instead of, or in addition to, their label text.
    • Quick Reply Sets can now be assigned colors to highlight the buttons.
    • Added a basic debugger to the Quick Reply editor.
    • Added UI shortcuts for copying or moving quick replies between sets.
    • Added an option to auto-execute QR on new chat creation.

STscript

The piped value from a parent closure will no longer be automatically injected into the first command of a child closure. You can still explicitly reference the parent's piped value with {{pipe}}, but if you leave the unnamed argument of the first command inside a closure blank, the value will not be automatically injected.

Messages added by Image Generation and Image Captioning extensions now emit events that can trigger QR auto-execution.

  • Added support for block comments.
  • /run command: can now execute closures directly, in addition to QRs and closure variables.
  • /break command: added to break out of a loop (/while or /times) or a closure early.
  • /addswipe command: added switch parameter to display a new swipe immediately.
  • /api-url command: added to get/set server URL for compatible APIs.
  • /tokenizer command: added to choose a preferred tokenizer.
  • /gen command: now supports empty prompts.
  • /api and /summarize now support quiet parameter to hide toast notifications.

STscript documentation: https://docs.sillytavern.app/usage/st-script/

Bug fixes

  • Fixed Google Translate API response decoding. No more ��.
  • Fixed being unable to toggle DRY sampling block with sampler select.
  • Fixed rendering of underscore italics Markdown in code blocks.
  • Fixed expanded custom CSS editor not applying changes to the styles.
  • Fixed OpenRouter providers selection not being loaded on app start.
  • Fixed sluggish performance of WI editor when editing large books.
  • Fixed lazy loading of chat when all visible messages were deleted.
  • Fixed using legacy API paths for sources where it was not supported.
  • Fixed automatic TTS narration of messages hidden from prompts.
  • Fixed a placeholder TTS text causing ghost narration on mobile.
  • Fixed vector storage applying summarization to files and WI entries.
  • Fixed entity name prefix trimming even if name display is preferred.
  • Fixed macros not being replaced in DRY penalty sequence breakers.

Community updates

New Contributors

Full Changelog: 1.12.4...1.12.5

Check out latest releases or
releases around SillyTavern/SillyTavern 1.12.5

Don't miss a new SillyTavern release

NewReleases is sending notifications on new releases.

Get notifications