Playground
Three small specialised models, trained on public data. Write a request: the engine decides and shows its confidence.
Your text is processed in memory, then forgotten: it is neither stored nor logged.
Banking intents
Classifies a customer request into one of 77 intents. Trained in English.
0 / 500 characters
Decision
Write a request or pick an example, then “Decide”.
- Latency of this decision
- —
- Median latency (server measure)
- ~35 ms
- Model size
- 294 MB
- Hardware
- 4 vCPU, no GPU
- Training data
- Banking77, CC BY 4.0
- Accuracy (public test)
- en 0.924
The same engine, on your data?
The diagnostic measures what a specialised model would achieve on your decision, before any commitment.
Datasets
Banking77: Casanueva et al., “Efficient Intent Detection with Dual Sentence Encoders”, PolyAI, 2020. CC BY 4.0 licence.
MASSIVE: FitzGerald et al., “MASSIVE: A 1M-Example Multilingual Natural Language Understanding Dataset with 51 Typologically-Diverse Languages”, Amazon, 2022. CC BY 4.0 licence.
Changes: we use the English, French, German and Italian parts of MASSIVE; the yes / no question is derived from MASSIVE (“calendar” scenario versus all other scenarios). The models are trained on these data; the data are not redistributed here. CC BY 4.0 licence text