Skip to content

Latest commit

 

History

History
63 lines (43 loc) · 3.07 KB

CHANGELOG.md

File metadata and controls

63 lines (43 loc) · 3.07 KB

Change Log

2023-05-31

  • Updated parameters and streaming for llama.cpp api server.
  • Changed to send the seed parameter when its value is >= 0.
  • Disabled searching and trimming stopping strings for backends that already do it well.
  • Added workaround for a SillyTavern extension that tries to send and parse a custom prompt.

2023-05-28

  • Added streaming and stopping strings for Kobold. Made the seed setting work on Kobold too.
  • Added an option to stream by character, it's enabled by default.

2023-05-21

  • Added initial support for llama.cpp API server. The API isn't very mature yet. There's a llamaCppSettings field in the config for additional parameters.
  • Added the option to always keep the first example message for the AI's character.

2023-05-14

2023-05-10

  • Added a default impersonation prompt in the config.
  • Fixed character bias when using pseudo-streaming.

2023-05-08

  • Reorganized the code structure.
  • Added option to disable sending the CORS headers.
  • Added an option to set the seed for ooba.
  • Added bluemoon format.
  • Added vicuna-cocktail format.
  • Added an alternative verbose format.
  • Added basic superbig support. There's a new config option and a python script at src/basic-superbig-api.py. It's probably easier to do "pip install superbig" in ooba's env and run the script inside it.
  • A custom impersonation prompt can be added on the second line after "IMPERSONATION_PROMPT".

2023-05-04

  • Added to abort the request when the stop streaming button is pressed.

2023-05-03

  • Added to cancel the previous request, at least when using koboldcpp and the horde, and wait for the previous request to finish before starting the new one.
  • Fixed Horde anonymous requests.
  • Added startup scripts.
  • Reverted "add support to set the character names in the main prompt." That prompt is not sent when using impersonation. Changed it back to the first line of the Jailbreak.
  • Added an option to include the character bias in the final text generated. It's enabled by default.
  • Fixed how the singleline prompt format finds who sent the last message and added an option to customize the "[says nothing]" message.

2023-05-02

  • Added Horde support, see config.default.mjs.
  • Added character bias (a string added at the very end of the prompt)
  • Added different configuration variable to set the max amount of tokens to generate while using impersonation.
  • Added support to set the character names in Main Prompt in the first line with this format "{{char}}|{{user}}", freeing the jailbreak. The following lines after the first one can be used normally.

2023-04-29

  • Added a config.mjs file for the settings.
  • Added presets/ for generation presets and prompt-formats/ for the functions that generates the prompts.