1.12.8

SillyTavern/SillyTavern

版本发布时间: 2024-12-01 23:33:05

SillyTavern/SillyTavern最新发布版本:1.12.9(2024-12-07 21:13:23)

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

New Contributors

@Beinsezii made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3017
@bal-spec made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3023
@dylan1951 made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3024
@procule made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3036
@yar3333 made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3037
@LuminousPath made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3068
@KaraKaraWitch made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3077
@poipoi300 made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3083
@Ivruix made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3092
@tincansimagine made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3100
@Finadil made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3103
@kallewoof made their first contribution in https://github.com/SillyTavern/SillyTavern/pull/3090

Full Changelog: https://github.com/SillyTavern/SillyTavern/compare/1.12.7...1.12.8

相关地址：原始地址下载(tar) 下载(zip)

查看：2024-12-01发行的版本