ZeroPoint

ZeroPoint chooses the option that fits a text from a list of options, or answers that none of them does. It runs on a CPU in about 20 ms, and the model ships inside the package.

Typical uses are routing a support ticket to a queue, finding the intent behind a request, or sending a document to the right process. A classifier always picks something. ZeroPoint returns None when the right option is not on the list or the text has nothing to do with the task.

Install

pip install zeropoint

ZeroPoint needs Python 3.10 or newer and runs on Linux, macOS and Windows, with ONNX Runtime. The package holds the model (89 MB); nothing is downloaded when it runs. PyTorch is only needed to adapt it to your labels.

First decision

from zeropoint import ZeroPoint

zp = ZeroPoint()
zp.decide("Can you move 200 euros from my savings to my checking account?",
          ["transfer", "balance", "pay bill"])

transfer  0.77  ███████████████·····
balance   0.02  ····················
pay bill  0.01  ····················
none      0.20  ████················
→ transfer · 17 ms

zp.decide("Set an alarm for 6:30 tomorrow morning", ["transfer", "balance", "pay bill"])

pay bill  0.00  ····················
transfer  0.00  ····················
balance   0.00  ····················
none      1.00  ████████████████████
→ none · 16 ms

Where to go next

Page	What it covers
Making decisions	options, what a decision holds, thresholds, many texts at once, languages
Adapting to your labels	fit and recalibrate on your own labelled texts
Running in production	threads, long texts, the same numbers on every machine, offline use
API reference	every class, method and field
Training data	what the model was trained on, and under which licences
Benchmark	ZeroPoint against Laya, Semantic Router and zero-shot NLI