DistilBART CNN 12-6 — ONNX

ONNX export of DistilBART CNN 12-6 (306M parameters) with encoder-decoder architecture and KV cache support.

A distilled version of BART Large CNN, fine-tuned for text summarization on the CNN/DailyMail dataset.

Converted for use with inference4j, an inference-only AI library for Java.

Original Source

try (var summarizer = BartSummarizer.distilBartCnn().build()) {
    System.out.println(summarizer.summarize("Long article text..."));
}

Property	Value
Architecture	BART encoder-decoder (306M parameters, 12 encoder + 6 decoder layers)
Task	Text summarization
Training data	CNN/DailyMail
Tokenizer	BPE (50,265 tokens)
Original framework	PyTorch (transformers)
Export method	Hugging Face Optimum (encoder-decoder with KV cache)

This model is licensed under the Apache License 2.0. Original model by Sam Shleifer.