Description
Within many character cards found around the internet, there are often a series of keywords associated with a snippet of text.
I find these extremely useful for reducing token usage in a card. For example, in a herbalist character card with a lot of complex data, not all of it needs to be loaded into the personality or description, most of the information can simply be stored within a number of Lorebook entries.
Personally, I enjoy creating small and helpful character cards, such as an apothecary, filled with a library of keyword entries like "headache," "headaches," "migraine," "stomachache," etc., each linked to proven and helpful snippets of information to guide the LLM.
So having the personality and description to be very small, and to have the bulk within the lorebook entries, its just a style of character card that I like to use and find helpful.
Essentially, I like to create survival cards. And when used with a mini LLM (since small LLMs running locally tend to struggle with real survival knowledge), I believe Lorebook entries will prove to be a very useful feature to complement this already amazing platform.
I understand that someone else has already requested the addition of Lorebooks, but I humbly ask that you reconsider your decision. Of course, I will respect whatever outcome you choose.
Proposed Solution
For a given lore item, if the key (case insensitive) appears in the recent chat history, the corresponding value will be included in the context when sending the message to the model. One can use wildcards to match multiple suffixes (e.g. "name*") or prefixes (e.g. "*name") in the key.
Description
Within many character cards found around the internet, there are often a series of keywords associated with a snippet of text.
I find these extremely useful for reducing token usage in a card. For example, in a herbalist character card with a lot of complex data, not all of it needs to be loaded into the personality or description, most of the information can simply be stored within a number of Lorebook entries.
Personally, I enjoy creating small and helpful character cards, such as an apothecary, filled with a library of keyword entries like "headache," "headaches," "migraine," "stomachache," etc., each linked to proven and helpful snippets of information to guide the LLM.
So having the personality and description to be very small, and to have the bulk within the lorebook entries, its just a style of character card that I like to use and find helpful.
Essentially, I like to create survival cards. And when used with a mini LLM (since small LLMs running locally tend to struggle with real survival knowledge), I believe Lorebook entries will prove to be a very useful feature to complement this already amazing platform.
I understand that someone else has already requested the addition of Lorebooks, but I humbly ask that you reconsider your decision. Of course, I will respect whatever outcome you choose.
Proposed Solution
For a given lore item, if the key (case insensitive) appears in the recent chat history, the corresponding value will be included in the context when sending the message to the model. One can use wildcards to match multiple suffixes (e.g. "name*") or prefixes (e.g. "*name") in the key.