The Dataset Viewer API answers what is in a split, what the columns look like and whether the text you need is actually in there - so the decision gets made from a handful of calls instead of a clone.
A server over the Hugging Face Dataset Viewer API. It reads datasets on the Hub without cloning them: what configurations and splits exist, paginated rows, per-split statistics, and search and SQL-style filtering over the contents.
- A dataset's detail - its configurations and its splits - before anything is downloaded
- The first rows of a split, and any page of it after that
- Statistics for a split, so the distribution is visible without loading the data
- Text searched across a split, and rows filtered with a SQL WHERE clause and an ORDER BY
- The whole dataset pulled down in Parquet, once you have decided it is the right one
- Private datasets reachable by passing a token on the call
Python 3.12 or newer. Public datasets need no credential; a private one needs a Hugging Face token, passed per call as auth_token rather than configured once. Datasets are addressed by their Hub identifier - stanfordnlp/imdb, for example.
Build from source — clone the repository and build it, then point your client at the binary