Daniel van Strien
@vanstriendaniel
Machine Learning Librarian @huggingface ๐ค
I like big datasets and small models.
1.5K Following 6.7K Followers
Trained a Jev-style classifier on @huggingface Jobs for ~$1.50.
It's a 194M GLiNER2 model that suggests task tags for any Hub dataset from its column names and first row, and returns a label with a probability.
Zero-shot, GLiNER2's first suggestion matched an owner's tag 10% of the time. After 17 minutes of fine-tuning: 69%. The fine-tuned model runs on a free CPU in about a second.
Owners' tags are noisy, so some "wrong" answers are tags the owner left out.
The recipe is open: one hf jobs command trains the same kind of model on your own labels. The README example (book titles) runs in ~2 minutes for about $0.02.
Demo:
Recipe:
Show more
1,274 hours of egocentric video with paired 7-IMU arm tracking, now CC-BY-4.0 on @huggingface.
@eidon_ai, a robotics company, is winding down and has released all their data: people doing laundry, cleaning, dishes and cooking. 13,451 recordings, 9 TB.
Show more
The real tokenmaxxing move: use frontier agents to build small classifiers for large-scale data curation.
Small classifiers already help curate the data used to train LLMs. I wanted to see how far an agent could take me in building one.
I used Astra, SetFit and @huggingface Jobs to turn 200 agent-labelled examples into a reusable document-purpose classifier, reviewing the categories and tricky cases along the way.
It classified 191,724 FinePDFs-Edu documents for ~$0.70 in inference compute, versus an estimated $13โ26 to label the same document excerpts with low-cost batch LLMs.
Training experiments added ~$2.90 in compute.
Workflow, mistakes, reusable model and a prompt to try on your own data:
Show more
Uploaded a dataset of 115,293 illustrated pages from the Encyclopaedia Britannica, 1st edition (1768) to 14th (1929) to the Hub
datatrove โ the data-processing library behind FineWeb, FineWeb2, and FinePDFs โ just shipped 0.10.0!
- JobsPipelineExecutor: run pipelines on @huggingface Jobs: fan-out, multi-stage dependencies, retries, resume. No Slurm cluster needed
- HF storage buckets as a DataFolder: read, write, and log straight to hf://buckets/...
- Reasoning outputs preserved in inference results
pip install datatrove[io,processing]
Show more
@liquidai AI just released LFM2.5 encoders small bidirectional models with 8k context.
I've added a training recipe for fine-tuning them on @huggingface Jobs (works for any encoder), and used it to train a classifier that suggests task categories for Hub datasets from their READMEs.
One command, 20 minutes on an A100, about $0.80.
Show more
Coding agents are real users of the Hub now i.e. Claude Code alone is ~24% of attributed agent traffic.
But many agents use the Hub badly: choose models from a year-old training cutoff, guessed CLI flags, no GPU.
Some tips to get agents to use @huggingface better ๐งต
Show more
Coding agents are real users of the @huggingface Hub!
They're searching for models, building and pushing datasets, training models on Jobs, spinning up Spaces...
Now there's public data: each agent's share of Hub traffic, updated monthly ๐
Show more
