AI Question Answering from a Passage (DistilBERT SQuAD)
Ask a question about a passage and get the exact answer span highlighted, from a neural model that runs in your tab.
Use the tool
Why this tool exists
Extractive question answering finds the stretch of text that answers a question, instead of generating new text. That makes it far less prone to invention than a chat model: the answer is always words from your passage, so you can check it at a glance. It suits contracts, manuals, meeting notes and articles you do not want to upload.
The model is distilbert-base-cased-distilled-squad (Apache-2.0, trained on SQuAD v1.1) with int8 weights and a hand-written JavaScript engine. Long passages are read in overlapping 384-token windows and the best span is returned. It cannot say "not in the text": a low confidence figure is your warning.
How to use it
- Paste the passage and type a question.
- Press Find the answer; the first run downloads 67 MB once.
- Read the highlighted span and the confidence; treat low confidence as "probably not in the text".
Frequently asked questions
Does it write answers?
No, it only selects a span from your text. It cannot summarise, combine facts from two places or do arithmetic.
How do I know the answer is right?
Read the highlighted sentence in context and check the confidence. Under about 20% usually means the passage does not contain the answer.
How long can the passage be?
Several pages. Longer text is split into overlapping windows, which takes longer; very large documents are better split by section.