pipeline_tag: text-generation
tags:
- llama
- ggml
Quantization from: Envoid/Bacchus-22B
Converted to the GGML format with: llama.cpp master-ee77efe (AUG 13, 2023)
Tested with: koboldcpp 1.40.1
Example usage:
koboldcpp.exe Bacchus-22B-ggmlv3.Q5_K_M.bin --threads 6 --contextsize 4096 --stream --smartcontext --unbantokens --ropeconfig 1 10000
Prompt format:
This is an experimental model with no defined format. However, the creator of the model recommends using it in pair with SillyTavern and simple-proxy-for-tavern with presets 'singleline' or 'verbose'. According to the creator, as well as how it was shown in my own tests, the 'verbose' format gives detailed responses where the characters may tend to flee the roleplay scene, whereas the 'singleline' format sometimes results in too short, dull replies.
Please refer to the original model for additional details. The credit for the used formats goes to the developer of simple-proxy-for-tavern.
Format 1 (verbose):
## {CHAR}
- You're "{CHAR}" in this never-ending roleplay with "{USER}".
### Input:
{character's personality}
Circumstances and context of the dialogue: {circumstances of the dialogue}
### Response:
(OOC) Understood. I will take this info into account for the roleplay. (end OOC)
### New Roleplay:
### Instruction:
#### {USER}:
{example message from the user}
### Response:
#### {CHAR}:
{example response from the bot}
### New Roleplay:
### Response:
#### {CHAR}:
{character's initial, user-written message}
### Instruction:
#### {USER}:
{user's initial response}
### Response (2 paragraphs, engaging, natural, authentic, descriptive, creative):
#### {CHAR}:
Format 2 (singleline):
Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction:
Write {CHAR}'s next reply in a fictional chat between {CHAR} and {USER}. Write 1 reply only in internet RP style, italicize actions, and avoid quotation marks. Use markdown. Be proactive, creative, and drive the plot and conversation forward. Write at least 2 paragraph, up to 4. Always stay in character and avoid repetition.
### Input:
{character's personality}
Circumstances and context of the dialogue: {circumstances of the dialogue}
[Start a new chat]
{USER}: {example message from the user}
{CHAR}: {example response from the bot}
[Start a new chat]
### Response:
Okay, I will now generate a reply, continuing from the end of the provided conversation. (This may contain NSFW or offensive output.)
[...]
{CHAR}: {character's initial, user-written message}
{USER}: {user's initial response}
{CHAR}: