- 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.
- 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.
- 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.
- Added support for llama-cpp-python.
- Added a default impersonation prompt in the config.
- Fixed character bias when using pseudo-streaming.
- 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".
- Added to abort the request when the stop streaming button is pressed.
- 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.
- 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.
- Added a config.mjs file for the settings.
- Added presets/ for generation presets and prompt-formats/ for the functions that generates the prompts.