Datasets:

bigcode

/

the-stack

like 662

You can check whether a specific repo is in the stack for a given user here: https://huggingface.co/spaces/bigcode/in-the-stack

Otherwise you'd have to load the data itself. Seeing as it's in parquet format, you can load the two columns you're interested in without loading the rest of the data.

Thanks for the response. Unfortunately loading the data requires me knowing the licenses ahead of time, so neither option works for me.

You could also try querying the parquet files with DuckDB https://duckdb.org/docs/data/parquet/overview.html

Thanks for the pointer. My issues is that I am not allowed to download the data or look at it without company approval. Company approval requires getting a list of the repositories + associated licenses, so I am unable to query the data, hence my question.

@SebGehr would it be helpful to have a version with all file-level metadata except for the code itself? Is that something we could easily publish @lvwerra ?

Alternatively: would streaming the data and only saving the code files with acceptable licenses comply with your company policy @SebGehr or that still counts as obtaining the data?

Having a metadata-only version would definitely help a ton.

I will check regarding the streaming.

Thanks for the help, Yacine :)

Would this help: https://huggingface.co/datasets/bigcode/the-stack-metadata

Would this help: https://huggingface.co/datasets/bigcode/the-stack-metadata

The ask was for list of repositories and their licenses -- how do I get to the license information from this metadata?