Predict masked words in modern Hebrew with DICTA's state-of-the-art Hebrew BERT model. Use the playground to give DictaBERT a representative input and review its output. Adjust the available settings to shape the result for your task.
Model
dicta-il/dictabert
Documentation
Use this reference to understand how every field in the playground changes the request sent to dicta-il/dictabert.
Required
Type: Text
Provides one Hebrew sentence or a JSON list of sentences containing exactly one [MASK] token each. The model ranks vocabulary tokens that could replace the mask, so surrounding context and correct Hebrew spelling directly influence the predictions.
Optional
Type: Integer
Default: 5
Range: 1–Unbounded
Sets how many highest-scoring replacements are returned for each masked token. Increase it to inspect more alternatives, or reduce it when only the most likely completions matter; it does not change the underlying prediction scores.
Optional
Type: JSON array
Default: None
Restricts masked-word prediction to the supplied JSON list of candidate words or tokens. Use it to compare a known vocabulary rather than search the entire model vocabulary; candidates that split into multiple tokens may not behave like a single masked-token replacement.
Optional
Type: JSON object
Default: None
Passes advanced JSON options directly to the tokenizer, such as truncation or maximum input length. Use it when the exposed controls do not cover the required preprocessing behavior, and avoid settings that remove the mask token or important source text.