1.12.8

[Cohee1207]

SillyTavern 1.12.8

Important News

• SillyTavern now uses Webpack for bundling frontend dependencies. This allows us to simplify dependency management and minimize library vendoring. The bundling process is performed in runtime when the server is started and may take up to 30 seconds to complete on a first run, the subsequent runs will be faster as the dependencies are cached. We made sure to minimize any performance overhead, but please report any issues you encounter.
• Formula rendering has been moved to an official downloadable extension and now uses a different syntax. You may have to use a regex polyfill to make it work with your existing formulas. See the documentation for more information: Formulas Rendering.
• SillyTavern can now be hosted using Deno and Bun. This has not been extensively tested or recommended for daily use, and compatibility with server plugins is not guaranteed. We will continue to support Node.js as the primary runtime for SillyTavern. Use npm run start:deno and npm run start:bun to start the server with Deno and Bun, respectively.
• If you have previously patched the AllTalk TTS provider manually, you will have merge conflicts on the next git pull. Refer to the documentation on how to resolve them: Common Update Problems.

Backends

• Added NanoGPT as a Chat Completion provider.
• Added new models for Claude, Cohere, MistralAI and Google AI Studio to the list of available models.
• Claude: The prompt converter no longer forces user messages at the beginning of the prompt.
• Claude: Now supports caching of the chat history. Refer to the community guide for more information: Claude Prompt Caching.
• KoboldCpp: Now supports logprobs.
• MistralAI: Now supports frequency and presence penalties.
• Aphrodite: Added DRY, skew, no repeat ngram and nsigma samplers.

Improvements

• Text Completion: Context size and built-in Advanced Formatting templates can now be derived from backends that implement the /props endpoint (llama.cpp and KoboldCpp).
• Text Completion: Added formatting templates for Mistral V7 and Tulu.
• Chat Completion: Prompt post-processing converters for Custom type now support multimodal image inlining.
• Chat Completion: Empty logit bias is no longer sent to the backend to improve compatibility with "OAI-compatible" backends.
• Logprobs: You can now reroll from an entire or partial prefix to continue the generation.
• Group Chats: Added a 'manual' reply strategy.
• An in-memory cache for parsed character cards is now disabled on Android and limited to 100 MB on other platforms to decrease the memory footprint.
• Performance improvements related to Chat Manager and fuzzy search.
• Various localization fixes and improvements.

Extensions

• Image Generation:
  • Added Black Forest Labs as an API provider.
  • Added NanoGPT as an API provider.
  • ComfyUI (breaking): Avatar image placeholders now use plain base64 encoding instead of data URI.
  • ComfyUI: Model selection dropdown now includes GGUFs and UNets.
• Regex: Added an option to escape macro value replacements in Find Regex.
• Vectors: Added a progress indicator for Data Bank files ingestion.
• TTS: Official support for AllTalk V2 API.
• Quick Replies: QR sets assigned as context menus now respect the "hidden" flag.

STscript

• Variable-related macros ({{getvar}}, {{setvar}}, etc.) are no longer evaluated line-by-line, so you may use newlines in variable values.
• Commands related to Author's Note now have the note- prefix. Added a /note-role command.
• Added /upper and /lower commands for text case conversion.

Bug Fixes

• Fixed prefixes of example messages for Gemini without a system prompt.
• Fixed auto-resizing of input textarea on narrow screens.
• Fixed an issue when dates localization would produce invalid dates.
• Fixed logprobs using sentence trimming on reroll.
• Fixed parsing of numeric series in math commands when using {{pipe}} as an unnamed argument.
• Fixed the id parameter not being respected in /qr-context commands.

Community Updates

• Add Webpack to eliminate frontend vendoring by @Cohee1207 in Add Webpack to eliminate frontend vendoring #2996
• llama.cpp Enable dry w/ array convert by @Beinsezii in llama.cpp Enable dry w/ array convert #3017
• Update azure.js by @bal-spec in Update azure.js #3023
• Add NanoGPT chat completions provider by @dylan1951 in Add NanoGPT chat completions provider #3024
• Optimize chat manager logic by @P3il4 in Optimize chat manager logic #3029
• Totally not a new macro engine by @Cohee1207 in Totally not a new macro engine #3035
• Enhancement: Token Probabilities viewer with dynamic prefix adjustment via Ctrl-click by @procule in Enhancement: Token Probabilities viewer with dynamic prefix adjustment via Ctrl-click #3036
• Remove embedded images before send text to TTS by @yar3333 in Remove embedded images before send text to TTS #3037
• Added Claude 3.5 Haiku by @WolframRavenwolf in Added Claude 3.5 Haiku #3045
• Backport llama.cpp-related DRY fixes from staging by @Cohee1207 in Backport llama.cpp-related DRY fixes from staging #3051
• Exclude links to embedded images from translation by @yar3333 in Exclude links to embedded images from translation #3059
• Update ru-ru translation by @Yokayo in Update ru-ru translation #3053
• feat: Add preferences parameter to /searxng endpoint by @LuminousPath in feat: Add preferences parameter to /searxng endpoint #3068
• Supports GEMINI EXP 1114 by @M0ch0 in Supports GEMINI EXP 1114 #3073
• Make featherless.ai Search bar take up 1 entire row on mobile devices. by @KaraKaraWitch in Make featherless.ai Search bar take up 1 entire row on mobile devices. #3077
• Remove best_of for VLLM by @Beinsezii in Remove best_of for VLLM #3076
• Support for AllTalk V1 and V2 by @erew123 in Support for AllTalk V1 and V2 #3060
• Fixed typo in HTML title by @poipoi300 in Fixed typo in HTML title #3083
• Make dynamic reroll available without use of modifier key by @ceruleandeep in Make dynamic reroll available without use of modifier key #3078
• Enhancement: Make buttons scrollable by @joenunezb in Enhancement: Make buttons scrollable #3004
• Fix: Inconsistent Textarea resizing in small width windows by @joenunezb in Fix: Inconsistent Textarea resizing in small width windows #3006
• Feature/comfy gguf models by @ceruleandeep in Feature/comfy gguf models #3088
• Claude caching at depth by @Honey-Tree in Claude caching at depth #3085
• Add command-line arguments description to Russian readme by @Ivruix in Add command-line arguments description to Russian readme #3092
• Mistral-2411 context+instruct presets by @WolframRavenwolf in Mistral-2411 context+instruct presets #3093
• feat: Add multiple delayed model updates by @tincansimagine in feat: Add multiple delayed model updates #3100
• Add author's note STscript returns, role changing, and aliases by @Finadil in Add author's note STscript returns, role changing, and aliases #3103
• feat: Improve Korean data-i18n by @tincansimagine in feat: Improve Korean data-i18n #3098
• Derived templates by @kallewoof in Derived templates #3090
• Add 3-state find substitution select by @Cohee1207 in Add 3-state find substitution select #3105
• Fix "send a photo of yourself" etc in SD interactive mode by @ceruleandeep in Fix "send a photo of yourself" etc in SD interactive mode #3102
• Fix/conn refused err msg by @ceruleandeep in Fix/conn refused err msg #3099
• feat: add top nsigma sampler support by @AlpinDale in feat: add top nsigma sampler support #3094
• Template derivation: explicitly skip empty templates by @kallewoof in Template derivation: explicitly skip empty templates #3111
• misc: add DRY, skew, and no repeat ngram penalty for aphrodite by @AlpinDale in misc: add DRY, skew, and no repeat ngram penalty for aphrodite #3113
• Claude: remove user filler from prompt converter by @Cohee1207 in Claude: remove user filler from prompt converter #3108
• Add Tulu templates by @kallewoof in Add Tulu templates #3117
• tulu template: added \n to input/system_suffix to match model card by @kallewoof in tulu template: added \n to input/system_suffix to match model card #3118
• Make QR context menu display options more consistent with QR bar by @ceruleandeep in Make QR context menu display options more consistent with QR bar #3122
• Wire up id= parameter for /qr-context* by @ceruleandeep in Wire up id= parameter for /qr-context* #3121
• (perf) Add 100MB limit to parsed characters cache by @Cohee1207 in (perf) Add 100MB limit to parsed characters cache #3120
• Replace webpack-dev-middleware with a statically compiled file by @Cohee1207 in Replace webpack-dev-middleware with a statically compiled file #3119
• Filter out hidden items in context menus by @ceruleandeep in Filter out hidden items in context menus #3124
• perf(search): improve fuzzy character search performance by ~13x (4.5s → 350ms) by @joenunezb in perf(search): improve fuzzy character search performance by ~13x (4.5s → 350ms) #3106
• feature: allow auto-use of max context size given by backend by @kallewoof in feature: allow auto-use of max context size given by backend #3112

New Contributors

• @Beinsezii made their first contribution in llama.cpp Enable dry w/ array convert #3017
• @bal-spec made their first contribution in Update azure.js #3023
• @dylan1951 made their first contribution in Add NanoGPT chat completions provider #3024
• @procule made their first contribution in Enhancement: Token Probabilities viewer with dynamic prefix adjustment via Ctrl-click #3036
• @yar3333 made their first contribution in Remove embedded images before send text to TTS #3037
• @LuminousPath made their first contribution in feat: Add preferences parameter to /searxng endpoint #3068
• @KaraKaraWitch made their first contribution in Make featherless.ai Search bar take up 1 entire row on mobile devices. #3077
• @poipoi300 made their first contribution in Fixed typo in HTML title #3083
• @Ivruix made their first contribution in Add command-line arguments description to Russian readme #3092
• @tincansimagine made their first contribution in feat: Add multiple delayed model updates #3100
• @Finadil made their first contribution in Add author's note STscript returns, role changing, and aliases #3103
• @kallewoof made their first contribution in Derived templates #3090

Full Changelog: 1.12.7...1.12.8

---

This discussion was created from the release 1.12.8.