Browser-ready ONNX export of the cive202/humanize-ai-text-bart-large humanizer (406M, BART-large), converted by Lynote for 100% in-browser inference with transformers.js.

The export is built directly from the MIT-licensed upstream weights with torch.onnx (scripts/model/export-bart-split.py in StealthHumanizer): encoder_model.onnx + decoder_model_merged.onnx (incremental past-key-value cache), the exact layout transformers.js expects for seq2seq models.

Attribution

Upstream fine-tune: cive202/humanize-ai-text-bart-large (MIT)
Base model: facebook/bart-large (MIT)
Conversion: ONNX export by Lynote. Lynote does not claim authorship of the upstream weights.

Usage (transformers.js)

import { pipeline } from '@huggingface/transformers';
const humanize = await pipeline('text2text-generation', 'Lynote/cive202-humanize-ai-text-bart-large-browser');
const out = await humanize('It is important to note that this robust solution is seamless.', { max_length: 150 });

Or via the StealthHumanizer app with NEXT_PUBLIC_BROWSER_MODEL_BASE=https://huggingface.co/Lynote.

Responsible use

This is a writing-quality aid, not a guarantee that text will be classified as human by any AI detector. Do not use it to misrepresent authorship in academic, legal, or disciplinary contexts.