KurdishCorpus-clean — a related corpus, worth comparing notes?
#2
by alanhasn - opened
Hi Mevlan — came across your work on OpenKurdish (the Llama3 checkpoints
and Golden Corpus) after your Open Kurdish AI profile's contact email
turned out to be dead. Really glad to find someone else building in
this exact space.
We're Kurdish-Tech (github.com/Kurdish-Tech) — alongside a dictionary,
keyboard layout and spell-checker, we recently published
KurdishCorpus-clean (huggingface.co/datasets/kurdish-tech/KurdishCorpus-clean):
2.7M documents / 2.97B tokens across Kurmancî, Soranî and Zazakî,
deduplicated and quality-filtered.
Given how aligned our missions are, would be great to compare notes,
cross-link resources, or find some way to collaborate rather than build
in parallel without knowing about each other.