I found that when we have images in our docs the base64 text is often being included in the context (ex: when I was working on https://github.com/fastai/fastcore/pull/621). I think this might be taking up too many tokens, and maybe good to exclude them?
I found that when we have images in our docs the base64 text is often being included in the context (ex: when I was working on https://github.com/fastai/fastcore/pull/621). I think this might be taking up too many tokens, and maybe good to exclude them?