schift-nli

ONNX-quantized DeBERTa-v3-xsmall for Natural Language Inference, optimized for Dot local inference.

• Base model: cross-encoder/nli-deberta-v3-xsmall
• ONNX source: Xenova/nli-deberta-v3-xsmall
• Parameters: 70.8M
• Labels: entailment, contradiction, neutral
• Use case: Intent routing, polarity detection, document pair classification

Usage in Dot

Loaded automatically by Dot's local NLI classifier. No manual setup needed.

Usage with Transformers.js

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

const classifier = await pipeline('text-classification', 'schift-io/schift-nli', {
  quantized: true,
});

const result = await classifier(
  { text: 'A man is eating pizza', text_pair: 'A man is eating food' },
  { top_k: 3 }
);
// [{ label: 'entailment', score: 0.97 }, ...]

License

Apache 2.0 (inherited from base model)