NLP / MASKED LANGUAGE MODELING
Sentence Surgeon
Type any sentence. DistilBERT's WordPiece tokeniser splits it into the actual subword units the model sees. Click a token to mask it and watch the model predict what fills the gap.
Sentence
Tokens — click one to mask it
Tokens will appear once DistilBERT finishes loading.
The model is distilbert-base-uncased (~66 MB), a distilled BERT trained on English Wikipedia and BookCorpus, so its world model and biases reflect that. WordPiece is uncased, which is why predictions come back lowercased — and why rare words split into multiple ##-prefixed subwords.