← All models · Donut

use case · multi-model

Multi-model: answer → translate

One model can't do this alone. First Donut answers a question about the document; then M2M100 translates that answer into the language you pick. Read the paperwork, then make the answer readable to anyone — both models on your device.

Stage 1 — document VQA (Donut, ~215 MB)

Stage 2 — translation (M2M100 418M, ~470 MB)

Drop a document, or click to choose
An invoice A membership form A shop receipt
Selected document

1 · Answer (English)

2 · Translation

Why chain them? Donut turns a document into a fact; translation makes that fact readable across a language barrier — a private "read-this-form-for-me" tool where neither the document nor the answer ever leaves the device.

← Wild · Back to Donut