Commit Graph

4764 Commits

Author SHA1 Message Date
Cohee
85d25a8e13
Merge pull request #3060 from erew123/release
Support for AllTalk V1 and V2
2024-11-16 17:09:07 +02:00
Cohee
7c7aaf33fc
Merge pull request #3076 from Beinsezii/vllm_n_fix
Remove `best_of` for VLLM
2024-11-16 15:51:25 +02:00
Cohee
c9d2b609f1 match => includes 2024-11-16 15:41:41 +02:00
Cohee
37f4fd4def
Merge pull request #3073 from M0ch0/staging
Supports GEMINI EXP 1114
2024-11-16 15:40:56 +02:00
Cohee
33d8a91bf2 Linter fixes 2024-11-16 14:22:46 +02:00
Cohee
2dcdb4d14a Fix tags import 2024-11-16 14:22:23 +02:00
Beinsezii
c599f46269 Remove best_of for VLLM
It defaults to `n` anyways, and for some reason it was being incorrectly
received on 0.6.4.post1
2024-11-16 02:05:35 -08:00
erew123
e857db40fb
Updated AllTalk Extension to support AllTalk V2
1. Core Feature Addition:
   - Added AllTalk V1/V2 server version selection
   - Added RVC (Realistic Voice Conversion) support for V2
   - RVC features are automatically disabled when V1 is selected

2. Code Improvements:
   - Replaced custom debounce implementation with shared utils.js debounce utility
   - Fixed linting issues:
     - Converted HTML attribute quotes in template literals to single quotes
     - Added trailing commas where required
     - Fixed console.error message formatting

3. New Settings/Properties Added:
   ```javascript
   server_version: 'v2' (default)
   rvc_character_voice: 'Disabled'
   rvc_character_pitch: '0'
   rvc_narrator_voice: 'Disabled'
   rvc_narrator_pitch: '0'
   ```

4. Bug Fixes/Improvements:
   - Better error handling for RVC voice fetching
   - Improved URL handling for V1/V2 differences in API responses
   - Enhanced settings initialization and validation

5. Structural Changes:
   - Added RVC-specific UI elements and controls
   - Added version-specific logic for API endpoints
   - Improved settings synchronization between UI and backend

**NOTE** On line 70 there is an eslint bypass:

```javascript
        // HTML template literals can trigger ESLint quotes warnings when quotes are used in HTML attributes.
        // Disabling quotes rule for this one line as it's a false positive with HTML template literals.
        // eslint-disable-next-line quotes
        let html = `<div class="at-settings-separator">AllTalk V2 Settings</div>`;
```

The reason is:
1. ESLint's quotes rule wants all strings to use single quotes
2. However, this is a template literal containing HTML, where double quotes are standard for attributes
3. I tried various solutions:
   - Using single quotes: `<div class='at-settings-separator'>`
   - Using double quotes: `<div class="at-settings-separator">`
   - Even tried escaping quotes
   But ESLint just kept flagging it as an error
2024-11-15 13:14:38 +00:00
Cohee
9b38e3f449 Remove dated formula formatting references 2024-11-14 22:28:46 +02:00
M0cho
30bca8e39b Supports GEMINI EXP 2024-11-15 05:26:10 +09:00
Cohee
f1bda3fb22
Merge pull request #3053 from Yokayo/staging
Update ru-ru translation
2024-11-11 23:04:05 +02:00
Cohee
a9aff676ba [chore] Apply lint and formatter 2024-11-11 22:36:06 +02:00
erew123
3138252928
Support for AllTalk V1 and V2
Select the version in the interface. 
RVC is not supported on AllTalk V1
2024-11-10 20:25:08 +00:00
Yaroslav Sivakov
ba2b251d6e Exclude links to embedded images from translation 2024-11-10 17:30:55 +03:00
Cohee
b29bc4000c Fix character note for the first character in the list 2024-11-08 10:55:10 +02:00
Cohee
a3ca407b27
Backport llama.cpp-related DRY fixes from staging (#3051)
* llama.cpp Enable dry w/ array convert

The new PR that was merged needs an array instead of a str

https://github.com/ggerganov/llama.cpp/pull/9702

* Safe sequence breakers parse

* Support comma-separated list of llama.cpp sequence breakers #3026

---------

Co-authored-by: Beinsezii <beinsezii@gmail.com>
2024-11-06 19:47:30 +02:00
Yokayo
88ad22196c Merge branch 'staging' of https://github.com/Yokayo/SillyTavern into staging 2024-11-06 01:33:27 +07:00
Yokayo
9d664bc679 Update tl 2024-11-06 01:33:24 +07:00
Cohee
abef12d403 Fix "OpenAI-compatible" endpoints choking on empty logit bias 2024-11-05 16:42:51 +00:00
Wolfram Ravenwolf
dcd11e4c54 Added Claude 3.5 Haiku 2024-11-04 20:19:39 +01:00
Cohee
c8a28137e0 Fix numeric series parsing
Closes #3043
2024-11-04 12:28:38 +00:00
Cohee
77be125a99 NanoGPT: Add to image generation extension 2024-11-03 20:37:30 +02:00
Cohee
649c3911eb Add shouldRegister arg to /tools-register 2024-11-03 19:11:07 +02:00
Cohee
fe469745b3
Merge pull request #3037 from yar3333/release
Remove embedded images before send text to TTS
2024-11-03 19:02:44 +02:00
Yaroslav Sivakov
52a4c9202d
Update public/scripts/extensions/tts/index.js
Co-authored-by: Cohee <18619528+Cohee1207@users.noreply.github.com>
2024-11-03 19:57:33 +03:00
Yaroslav Sivakov
8b32ee9e80 Remove embedded images before send text to TTS 2024-11-02 20:25:25 +03:00
Cohee
8c3ac0ac5a Fix instruct macros not using a proper enabled field 2024-11-02 16:34:21 +02:00
Cohee
cfd1545b8a
Merge pull request #3036 from procule/update-token-probabilities
Enhancement: Token Probabilities viewer with dynamic prefix adjustment via Ctrl-click
2024-11-02 16:25:20 +02:00
Cohee
48c3e81f42 DRY 2024-11-02 13:22:41 +02:00
Cohee
293d6ff60d Run mutating variable macros first 2024-11-02 12:53:44 +02:00
Olivier Gagnon
abf7b90135 Enhance Token Probabilities viewer with selective reroll functionality and text extraction helper
- Updated onPrefixClicked() to add Ctrl-click functionality:
  - Default click rerolls the entire prefix text to continue the generation.
  - Ctrl-click rerolls only the text before the clicked word, allowing users to more precisely control generation output.
- Introduced getTextBeforeClickedWord() helper function:
  - Extracts the text before a clicked word based on click coordinates, enhancing control over prefix selection.
  - Provides a fallback to return the entire span text if word extraction is not possible.
- Updated function documentation to clarify new behaviors and helper purpose.

This change allows users to reset the prefix at a certain point if needed, adding flexibility to the generation process and enabling more intentional selection of alternatives.
2024-11-02 01:49:37 -04:00
Cohee
48d8e6e2c3 Port #3031 onto new engine 2024-11-02 00:55:34 +02:00
Cohee
5c90c8b1f6 Add post-process fn to evaluation 2024-11-02 00:44:12 +02:00
Cohee
8f373cf1dc Macros: refactor with a single replace point 2024-11-01 21:47:25 +02:00
Cohee
b837c482fc
Merge pull request #3029 from P3il4/staging
Optimize chat manager logic
2024-11-01 20:17:29 +02:00
Cohee
07feccbe35 koboldcpp: parse logprobs 2024-11-01 11:38:31 +02:00
Cohee
8c568bfa13 Merge branch 'staging' into optimize-chat-manager 2024-10-31 21:05:36 +02:00
Cohee
6e36b77f1a Remove TAI artifact 2024-10-31 19:57:29 +02:00
Cohee
f4ef324203
Merge pull request #3024 from dylan1951/add-nano-gpt-provider
Add NanoGPT chat completions provider
2024-10-31 19:30:24 +02:00
Cohee
4babf322c1 Fix model restoration on load 2024-10-30 02:09:28 +02:00
Cohee
f5f11eebb2 Support auto-connect 2024-10-30 02:02:56 +02:00
Cohee
085d852b57 Trigger inputs 2024-10-30 01:52:53 +02:00
Cohee
5ee0a6ec30 Support unlocked context size 2024-10-30 01:50:31 +02:00
dylan
e7522bba76 Populate model list from models endpoint 2024-10-29 19:38:46 +13:00
Cohee
00f0f755fc Support comma-separated list of llama.cpp sequence breakers #3026 2024-10-28 11:44:26 +00:00
Cohee
542f77aeb8 Safe sequence breakers parse 2024-10-28 11:39:59 +00:00
Cohee
517b140911 Localize only the moment instance 2024-10-28 11:07:39 +02:00
Cohee
9493d05f2c Localize only the moment instance 2024-10-28 11:01:48 +02:00
Beinsezii
ace2902cb8 llama.cpp Enable dry w/ array convert
The new PR that was merged needs an array instead of a str

https://github.com/ggerganov/llama.cpp/pull/9702
2024-10-26 16:07:07 -07:00
Cohee
f5bdb52c25 Merge branch 'staging' into webpack 2024-10-26 19:11:29 +03:00