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
fix(examples): example 04 called the entropy confidence "calibrated"
Aashish9112c8fbfd7c
diff --git a/examples/04_choice_question.py b/examples/04_choice_question.pyindex 8579d2dc..18fd5fbf 100644--- a/examples/04_choice_question.py+++ b/examples/04_choice_question.py@@ -10,7 +10,8 @@ banner("04", "Routing a support email (choice)", """A `choice` question gives the model a fixed set of labels. `criteria` is alabel -> description map: each key becomes one answer option and its descriptiontells the model what that label means. The answer then carries the chosen label, a- probability for every label, and a calibrated confidence.+ probability for every label, and two confidence numbers: `answer_confidence` (max(p),+ the number to gate on -- example 18 routes at a threshold on it) and `confidence`.`confidence` is not the top probability: it is the normalised entropy of the wholedistribution, 1 - H(p) / log(k), so it rewards one option clearly beating the rest.@@ -47,6 +48,7 @@ for label, state in CASES:answer = agent.predict(state, DEPARTMENT)["answers"]["department"]print(" chosen label : %s" % answer["choice"])print(" confidence : %.3f" % answer["confidence"])+ print(" answer_conf : %.3f" % answer["answer_confidence"])print(" all label probabilities:")for option, prob in answer["probabilities"].items():bar = "#" * int(round(prob * 32))