I noticed in https://rstudio.github.io/bslib/articles/value-boxes/index.md that the LLM-facing markdown files (which are really nice btw!) include base64-encoded images. Since models can't see these directly when they're included in text, they inflate the token count considerably without adding much value.
You can probably reuse remove_base64_images() from btw (added in posit-dev/btw#119)
I noticed in https://rstudio.github.io/bslib/articles/value-boxes/index.md that the LLM-facing markdown files (which are really nice btw!) include base64-encoded images. Since models can't see these directly when they're included in text, they inflate the token count considerably without adding much value.
You can probably reuse remove_base64_images() from btw (added in posit-dev/btw#119)