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(serve): skip torch-dependent tests when torch is not installed
Swatantra Yadav502fa2996803
diff --git a/tests/test_serve.py b/tests/test_serve.pyindex cc381dfc..5d24ab82 100644--- a/tests/test_serve.py+++ b/tests/test_serve.py@@ -245,6 +245,7 @@ def test_helpers():def test_thread_limit(monkeypatch):+ pytest.importorskip("torch")monkeypatch.delenv("LAYA_THREADS", raising=False)assert _apply_thread_limit() is None # unset -> no-op, no torch importfor bad in ("0", "-4", "abc", ""):@@ -421,6 +422,7 @@ def test_a_nested_choice_label_is_a_caller_error_not_a_server_fault(monkeypatch)`ValueError` to 422, so the caller got a 500 "inference failed" with the reason discarded.`ValueError` is what carries the message to the client, so the guard has to raise that type."""+ pytest.importorskip("torch") # laya.agent imports torch at module levelclass ValidatingRouter:"""The real guard, without a checkpoint: what `Agent.system_one` runs before encoding.