Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
docs(evals): both row routes refuse the same document, before the first call
Aashish66688f8fb834
diff --git a/docs/evals.md b/docs/evals.mdindex d9be6c21..f90c9df7 100644--- a/docs/evals.md+++ b/docs/evals.md@@ -273,6 +273,12 @@ One JSON object per line (JSONL). Blank lines and lines starting with `#` are ig`research/evals/dataset.template.jsonl` has a commented example.+A row built in Python with `Example(...)` is held to the same three rules as a JSONL row: `Dataset`+refuses one whose `questions` or `expected` is not an object, or whose `expected` names a question+`questions` does not carry, and quotes the row's index in the message. Both routes fail as+`EvalError` before the first model call, so a mistake in a dataset never arrives as a verdict about+the runner.+## MetricsEach metric is computed per answer where it applies and aggregated over the dataset: