As we know, the retrieval of files in Assistants involves the use of context tokens. However, in my efforts to make interactions more efficient and cost-effective, I’ve become interested in how assistants process instructions. Specifically, do these instructions also make use of context tokens?
My research indicates some instances where assistants with no file retrieval, seem to employ context tokens. This observation led me to wonder if perhaps fine-tuning a model to mimic these custom instructions could be a key strategy in achieving cost-effectiveness.