Key Info

A Hugging Face community member trained a GLiNER2 model for ~$1.50 that suggests task tags for any Hub dataset from column names and first row, with zero-shot accuracy of 10% improving to 69% after 17 minutes of fine-tuning.

Highlights

  • Cost: ~$1.50 for training; fine-tuned model runs on free CPU in about a second.
  • Recipe is open: one hf jobs command trains the same kind of model on your own labels; README example (book titles) runs in ~2 minutes for ~$0.02.
  • Demo and recipe links are provided in the original post.