About missing splits in dataset release
Hi, thanks so much for sharing this great resource and your excellent work!
I noticed that in the StarCoder2 paper (Table 4 at https://arxiv.org/pdf/2402.19173), the-stack-v2-train-extras includes 12 different sources. However, the splits Pull requests, Jupyter structured, and Jupyter scripts mentioned in the paper seem to be missing from the current dataset release.
Could you please clarify if these splits are available elsewhere or if I'm overlooking something?
Thanks again for your time!
Hi, these datasets are based on Software Heritage data and require an agreement with them like the one in The Stack v2. But you can find the code for building the datasets here https://github.com/bigcode-project/the-stack-v2/ we can also share s3 pointers to the data in SWH bucket like here https://huggingface.co/datasets/HuggingFaceTB/smollm-corpus#downloading-the-data
Gotcha. Thanks!