Assuming you already know what context in terms of LLMs, prefilling is the process of converting the current conversation into tokens and passing that into the LLM.