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): dataset template with the format
Emiliano Gandini Outeda04dd6b6c6b4e
diff --git a/research/evals/dataset.template.jsonl b/research/evals/dataset.template.jsonlnew file mode 100644index 00000000..17d3ccfc--- /dev/null+++ b/research/evals/dataset.template.jsonl@@ -0,0 +1,7 @@+# One JSON object per line. Blank lines and lines starting with # are ignored.+# Required: state, questions, expected (keyed by question id).+# Optional: tags (list of strings), language, model.+#+# The answer type follows the question: a label string for choice, a number for score,+# true/false for noul.+{"state": "I was charged twice, please refund me.", "questions": {"intent": {"type": "choice", "instructions": "What is the user asking for?", "criteria": {"billing": "refunds or charges", "technical": "bugs or errors"}}}, "expected": {"intent": "billing"}, "tags": ["billing"], "language": "en"}