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(nix): defaultModel is multilingual by default; english only for mostly English
Bruce-Yii347fc5696d47
diff --git a/nix/laya-serve.nix b/nix/laya-serve.nixindex 3ae5b63b..e6b261a0 100644--- a/nix/laya-serve.nix+++ b/nix/laya-serve.nix@@ -307,13 +307,13 @@ indescription = ''Checkpoint a state with no language evidence falls back to (sets`LAYA_DEFAULT_MODEL`): text with no letters at all, or Latin script too- short to identify a language. Set it to `multilingual` when most traffic- on this host is not English. Text the detector can place is routed on+ short to identify a language. Set it to `english` when most traffic+ on this host is English. Text the detector can place is routed onwhat it detects, so this is a fallback and not a pin. The spellingsaccepted, and the aliases among them, are laya's own -- the serverchecks the value against them at startup and refuses to start on a nameit does not know, so this module deliberately lists none. null leaves- laya's default, which is `english`.+ laya's default, which is `multilingual`.'';};