bluzen/tagbench-2026

tagbench-2026

Held-out tagging benchmark with per-tag frequency strata and a published error analysis for every baseline.

This card is generated from the entry's metadata in src/data/hub.js. Write a card field on the entry to replace it with a real one.

How to use it

from datasets import load_dataset

ds = load_dataset("bluzen/tagbench-2026", split="train")

Details

Limitations

Every dataset the lab publishes carries the biases of the corpus it came from, and ours is illustration-heavy. Measure before you trust it.