🎯 Train your own text classifier

Bring a labeled dataset, pick a base model, and get a fine-tuned classifier back. Runs on CPU by default; flip on ZeroGPU for a speed boost when it's available.

How it works: upload a file (.csv, .tsv, .json, .jsonl, .parquet) or type a Hub dataset (e.g. SetFit/sst2, fancyzhx/ag_news) → pick the text and label columns → hit Train. Labels can be words or numbers. Text-pair tasks (like NLI) work too via the optional second text column.

💡 On CPU, small models (DistilBERT) with ≤ a few thousand rows and 128 tokens work best.

1️⃣ Your data
Text column
Label column
Second text column (optional, for sentence-pair tasks)

Preview

2️⃣ Model & settings
Base model

Any Hub encoder model works; type a custom id if you like.
1 10
Batch size
32 512
0.05 0.5

⚡ Compute
200 50000
60 1800

Up to 120s of GPU; falls back to CPU if unavailable.