7 models · ranked by HuggingFace downloads
BART-large fine-tuned on MultiNLI for zero-shot text classification via natural language inference. Given a text and a candidate label, it predicts entailment probability to classify without task-specific training data.
nli-deberta-v3-base targets zero-shot text classification and is shipped as an open-weight, self-hostable checkpoint. Permissive Apache 2.0 terms let nli-deberta-v3-base go straight into commercial pipelines. Like most open checkpoints, nli-deberta-v3-base rewards a quick in-domain eval before commitment.
DeBERTa-v3-small fine-tuned on MNLI+SNLI for natural language inference and zero-shot text classification, from the sentence-transformers team. The small DeBERTa-v3 backbone outperforms BERT-base on NLI while being much faster than DeBERTa-v3-large.
distilbart-mnli-12-1 is an open-weight zero-shot text classification model in the bart family. Evaluate distilbart-mnli-12-1 on your own data before trusting it in production.
mDeBERTa-v3-base-xnli-multilingual-nli-2mil7 targets zero-shot text classification and is shipped as an open-weight, self-hostable checkpoint. Permissive MIT terms let mDeBERTa-v3-base-xnli-multilingual-nli-2mil7 go straight into commercial pipelines. mDeBERTa-v3-base-xnli-multilingual-nli-2mil7 is multilingual by design rather than English-only. Evaluate mDeBERTa-v3-base-xnli-multilingual-nli-2mil7 on your own data before trusting it in production.
As a distilbert-based open-weight model, distilbert-base-uncased-mnli focuses on zero-shot text classification. Read distilbert-base-uncased-mnli's card for hardware requirements and licensing fine print before deploying.
nli-MiniLM2-L6-H768 is an open-weight checkpoint for zero-shot text classification, distributed on the HuggingFace Hub. The Apache 2.0 license keeps nli-MiniLM2-L6-H768 unrestricted for commercial reuse. Like most open checkpoints, nli-MiniLM2-L6-H768 rewards a quick in-domain eval before commitment.