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
test(confidence): two `:NNN` citations spilled past _decode_answers
Nandakishor2bb5b08978b4
diff --git a/tests/test_confidence.py b/tests/test_confidence.pyindex ad2f5640..a424d8c7 100644--- a/tests/test_confidence.py+++ b/tests/test_confidence.py@@ -433,7 +433,7 @@ check("export/laya re-exports the same tuple", laya.GATE_STATES, _confidence.GAT# --------------------------------- the two doc pages must attribute confidence per question type# `docs/structured.md` and `docs/questions-and-answers.md` each opened their confidence section with# "`confidence` is normalized entropy". Neither agent writes that for every type:-# `Agent._decode_answers` (`laya/agent.py:1367-1404`) and `ONNXAgent._decode_answers`+# `Agent._decode_answers` (`laya/agent.py:1369-1398`) and `ONNXAgent._decode_answers`# (`laya/onnx_agent.py:682-727`) put `confidence_from_probs(p, k)` = `1 - H(p) / log(k)` in the# `choice` and `score` answers and `max(p_true, 1 - p_true)` in the `noul` one. Each page printed a# `noul` in the very block its sentence introduces, so the sentence was falsifiable from the page@@ -2228,7 +2228,7 @@ for _fn in (test_tl_kernels_drops_the_wrong_row_claim,# --------------------------------------- DecisionResult's docstring must split confidence by question type# `laya/structured.py`'s `DecisionResult` closed with "confidence keeps the **normalized-entropy# value** it has always had, because that is a different quantity on a scale that depends on the-# label count". `Agent._decode_answers` (`laya/agent.py:1349-1402`) uses two different formulas:+# label count". `Agent._decode_answers` (`laya/agent.py:1369-1398`) uses two different formulas:# * `choice` and `score` -> `confidence_from_probs(p, k)` = `1 - H(p) / log(k)`# * `noul` -> `max(p_true, 1 - p_true)` (its own comment: "identical here:# over two options `max(p_true, 1 - p_true)` is `max(p)`")