Clearing context window of LLM in Huggingface

Question

I want to use inference to ask different questions to LLMs taken from huggingface. But, I want to ask the prompts without the model having info about the previous prompts. Does the model automatically store the previous prompts in context?

Or does it not save any previous information at all and we need to provide all the context in the same prompt?

Answer 1

LLMs generally don't store your prompts or context or directly learn from it. After training LLMs stay static and the models weights don't change during inference. IF you want to build a chatbot you have to actively build something to keep the context. One solution for handling context is LangChain (https://github.com/langchain-ai/langchain).

But for your case, you just need to prompt to the LLM or the API to the LLM directly.