Fine-tuning a 194M GLiNER2 tag classifier on HF Jobs for ~$1.50 lifts accuracy from 10% to 69%
vanstriendaniel · x · 2026-09-23
The author trained a Jev-style classifier on Hugging Face Jobs for $1.50: a 194M GLiNER2 model that suggests task tags for any Hub dataset from its column names and first row. Zero-shot, its top suggestion matched an owner's tag 10% of the time; after 17 minutes of fine-tuning, 69% (owner tags are noisy, so some misses are unlabeled tags). It runs in 1s on a free CPU. The recipe is open: one hf jobs command trains the same model on your own labels, and the README book-titles example runs in 2 minutes for $0.02.
More from coding & agent
- Low-VRAM user claims applying a Qwen-style chat template fixes Ternary Bonsai 2's tool calling — needthosepylons · 2026-09-23
- pi-voice: Handy creator ships a voice extension for Earendil Pi — mitsuhiko · 2026-09-23
- Practical guide: using Gemini API skills with your coding agents — patloeber · 2026-09-23
- Redis ships Radar, Search on Flex and agent memory tools as 97% back context engineering — antirez · 2026-09-23
- shuding's essay 'Metaphors': protocols turn n×n integrations into n+n, and programs are proofs — shuding · 2026-09-23
- Jev's 32k Context Limit: Are People Misusing It for Full Agent Orchestration? — fastwind18 · 2026-09-23