How to Make Limited Memory For LLM

[tall matrix]

Hello I am trying to figure out how to program in limited memory to a LLM model is previous data/token optimized and fed back into the new model how is this done in a technical sense? I am extremely close to making a Virtual model myself the last two challenges are further optimizing my STT and making the memory for the API I have everything else set up even a cute new vtuber model 😄

Ehhh wish I could sent txt files would have asked if thats the right track

[uncut tartan]

This place is probably the poorest venue to get any answers, a lot of people theory-crafting only and script users, and no other actual devs with experience will answer you. The easiest is to use a library such as langchain/langflow:

https://python.langchain.com/docs/use_cases/chatbots

https://github.com/logspace-ai/langflow

Try joining actual dev groups if you want to learn proper stuff.

[thick thorn]

extract input embeddings, store into vector db, llm to summarize and append to system prompt

[civic rapids]

I know I’m not a dev but I saw this video on YouTube and didn’t know if it applies https://youtu.be/QQ2QOPWZKVc?si=68Q5e-0zJDQKkmyn