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(mcp): handle auto model in routing fallback
soumojit-D48d7eea12a8938
diff --git a/laya/mcp/tools.py b/laya/mcp/tools.pyindex b3fc6b74..0f9bea6c 100644--- a/laya/mcp/tools.py+++ b/laya/mcp/tools.py@@ -151,7 +151,15 @@ def laya_predict(# Router.predict and Agent.system_one both return the system_one payload,# which always carries an "answers" object (empty for empty questions).answers = _normalize_answers(result["answers"])- routing = result.get("routing") or {"model": model_name, "repo": None, "reason": "explicit model"}+ routing = result.get("routing")+ if not routing:+ # Agent.predict() doesn't return routing info; Router.predict() always does.+ # When model_name is "auto" (routing directive, not a checkpoint), we cannot+ # use it as the model name in the fallback.+ if model_name == "auto":+ routing = {"model": None, "repo": None, "reason": "auto routing without router"}+ else:+ routing = {"model": model_name, "repo": None, "reason": "explicit model"}# Real device of the checkpoint that answered: Agent.device reflects a# silent GPU -> CPU fallback. Omitted when it cannot be read, rather than# guessed.