affectively-ai/bart-large-cnn-onnx

BART Large CNN (ONNX, int8 Quantized)

Production-ready ONNX conversion of facebook/bart-large-cnn for in-browser text summarization — zero server cost, zero latency, complete privacy.

Highlights

• Abstractive summarization — generates concise summaries, not just extractive snippets
• ~493 MB quantized — BART-large architecture fine-tuned on CNN/DailyMail
• transformers.js compatible — drop-in pipeline('summarization')
• 406M parameters — high-quality summaries from long-form text

Quick Start

import { pipeline } from '@huggingface/transformers';

const summarizer = await pipeline(
  'summarization',
  'affectively-ai/bart-large-cnn-onnx',
  { dtype: 'q8' }
);

const result = await summarizer(
  'Today was an incredibly challenging day at work. I had back-to-back meetings...',
  { max_length: 50, min_length: 20 }
);
// [{ summary_text: 'The author had a challenging day with back-to-back meetings...' }]

Conversion Details

Property	Value
Base model	facebook/bart-large-cnn
Training data	CNN/DailyMail (300k news articles)
Export	PyTorch → ONNX via Optimum
Quantization	int8 dynamic (ORTQuantizer, avx512_vnni)
Quantized size	~493 MB

Use Cases

This model powers text summarization in Edgework.ai — bringing fast, cheap, and private inference as close to the user as possible. Best for:

• Summarizing journal entries for daily/weekly emotion reports
• Condensing therapy session notes
• Creating digests from long conversation threads
• Generating TL;DR for mental wellness content

About

Published by AFFECTIVELY · Managed by @buley

We convert, quantize, and publish production-ready ONNX models for edge and in-browser inference. Every release is tested for correctness and stability before publication.

• All models · GitHub · Edgework.ai