hfjev

Classify Hugging Face datasets across typed semantic dimensions with TypeSafe Jev System One.

npm install hfjev

Quick start

import hfjev from 'hfjev';

const dataset = await hfjev('cornell-movie-review-data/rotten_tomatoes');
const results = await dataset.classify();

console.log(results[0].answers);

hfjev() loads any Hugging Face dataset, auto-adapts evaluation rubrics to the domain, and classifies each row in a single parallel System One call with calibrated probabilities.

Gated & private datasets

const dataset = await hfjev('meta-llama/Llama-2-7b', {
  hfToken: process.env.HF_TOKEN,
  apiKey: process.env.TYPESAFE_API_KEY
});

Pass your Hugging Face User Access Token to authenticate gated or private datasets.

Custom dimensions

const dataset = await hfjev('ag_news');

dataset.adapt([
  { id: 'tech_relevance', type: 'noul', instructions: 'Is this about artificial intelligence?' },
  { id: 'urgency', type: 'score', instructions: 'Rate story urgency', criteria: ['Low', 'Breaking'] }
]);

const results = await dataset.classify();

adapt() overrides the default domain pack with your own Choice, Noul, or Score primitives.

Streaming evaluations

await dataset.stream((row) => {
  console.log(`[Row ${row.index}]`, row.answers);
});

stream() yields evaluations row-by-row for live feeds and telemetry without blocking on batch completion.

Local files and raw arrays

// Local JSON, JSONL, or CSV
const local = await hfjev('./reviews.json');

// In-memory array of strings or objects
const custom = await hfjev([
  'The acting was phenomenal throughout.',
  'Pacing dragged during the second act.'
]);

hfjev() detects intent directly from the input type.

CLI

npx hfjev cornell-movie-review-data/rotten_tomatoes --limit 5

Runs classifications directly from your terminal and prints formatted dimension scores and probabilities.

License

MIT © Hemanth.HM