hjLabs AI Playground
Starting the on-device runtime…
Fill-mask is the pre-training objective behind BERT and the clearest window into what a language model has actually learned: hide a word, and the model ranks candidates with probabilities. It is genuinely useful for probing bias, testing tokenisation, checking whether a domain term is in-vocabulary, and teaching how transformers work. Type a sentence with [MASK] in it and the top predictions appear with their scores, computed on your GPU.
fill mask demobert masked language modelpredict missing word aimlm playgroundbert bias probingtransformer demo online
BERT-family models use [MASK]; RoBERTa-family models use <mask>. The page inserts the correct token for the model you have selected, so you can click rather than remember.
It is a fast way to audit a model before you fine-tune it — whether your jargon survives tokenisation, and what stereotypes the pre-training data baked in.
No. The model weights are downloaded from Hugging Face to your browser once, and every inference after that runs on your own GPU through WebGPU. Your text, images and audio are never sent anywhere — hjLabs.in has no inference server and no way to see your input.
Small models do. Recent mobile Chrome and Safari expose WebGPU, but memory is the binding limit — stay under roughly 200 MB of weights on Android, and expect iOS to reclaim the tab sooner. The catalog shows every download size up front so you can pick something that fits.
That depends on the individual model, not on this site. Each model page links to its Hugging Face repository, where the licence is stated — most of the catalog is Apache-2.0 or MIT, a few are research-only. We add nothing on top: hjLabs.in claims no rights over anything you produce here.