AI Tools.

Search

text classification by FacebookAI

roberta-large-mnli

RoBERTa-large fine-tuned on the Multi-Genre Natural Language Inference (MNLI) corpus, commonly used for zero-shot text classification via the NLI entailment trick. One of the most frequently used models for zero-shot classification before dedicated models like DeBERTa-MNLI improved further.

Summary text generated by an automated pipeline from the model card · Not individually reviewed or run by us · How this page is made

From the model card

Fields below are copied from the tags and counters on the HuggingFace repository FacebookAI/roberta-large-mnli at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.

Publisher (HF namespace)
FacebookAI
Pipeline tag
text-classification
Library
Transformers
Framework tags
PyTorch, TensorFlow, JAX
Weight formats
safetensors
License tag
mit — read the license file in the repo before relying on it
Language tags
English (en)
Papers cited
arXiv:1907.11692, arXiv:1806.02847, arXiv:1804.07461, arXiv:1704.05426, arXiv:1508.05326, arXiv:1809.05053, arXiv:1910.09700
Datasets declared
multi_nli, wikipedia, bookcorpus
Downloads (HF counter at last fetch)
349,227
Likes (HF counter at last fetch)
210
Model card
https://huggingface.co/FacebookAI/roberta-large-mnli

Use cases

  • Zero-shot text classification via hypothesis entailment scoring
  • Document intent or topic classification without labeled data
  • Textual entailment for fact-checking pipeline components
  • Transfer learning baseline for NLI benchmark tasks

Pros

  • RoBERTa-large provides strong NLI quality for zero-shot classification
  • Well-established baseline with extensive literature comparison
  • Easy integration via pipeline('zero-shot-classification') in transformers
  • MIT license

Cons

  • DeBERTa-v3-large-mnli outperforms it on most zero-shot benchmarks
  • Zero-shot via NLI is slower and less accurate than a trained classifier when labels are available
  • Large model size (355M) for an encoder — slower than BERT-base alternatives
  • Multi-genre NLI training may not generalize well to domain-specific text

Tags

transformerspytorchtfjaxsafetensorsrobertatext-classificationautogenerated-modelcardendataset:multi_nlidataset:wikipediadataset:bookcorpusarxiv:1907.11692arxiv:1806.02847arxiv:1804.07461arxiv:1704.05426arxiv:1508.05326arxiv:1809.05053arxiv:1910.09700license:mit