Dataset card fields and split sizes differ from the Viewer

#2
by nwhite-systems - opened

The live Viewer currently exposes intent, text, spans, target and example_id, with train, validation and test splits. For example, the amh subset shows 2.24k / 320 / 640 rows. The Dataset Structure section instead documents question and answer fields and GSM8K-style train/test splits of 8 / 250 rows. Could these sections have been inherited from an earlier template? Aligning them with the published schema and splits would make loading and evaluation clearer for downstream users.

Sign up or log in to comment