I see. Thanks for the explanation internet friend!

Lots of ways to get code into LLMs, but this is one of my favorites for quick big experiments. I'm often using it in combo with Gemini's 2M window, testing limits on context size

Not "random", but now the output is trying to include images as SVGs or something because I raised the slider?

If so maybe the interface could warn or offer an option to exclude assets? I was hoping to capture oddly long contribution docs, examples, etc, but not media.

Ah! Then here it is - I thought this slider meant it would simply include random files in the repo that were unusually large. When I up the slider, that idea is confirmed by seeing more estimated tokens... I thought there was just more data available. What does the slider do? 🤷‍♂️

Yup yup. In this repo's output, I removed 6 occurrences, which took it from 2.7M tokens to about 400k: github.com/roboflow/sup...

I absolutely love this project - not a coder so using it all the time to provide context :)

One thing I've hit a couple times is that if the repo has a data outputting image/png the gitingest includes the whole hash. This quickly skyrockets the number of tokens, but I'm not sure how to exclude it.

excellent, thank you!

Is there a list of labelers somewhere? Didn't even know this existed til your post.