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
build(deps): cap NumPy below 2 in the onnx extra
thread3d495b590be438
diff --git a/pyproject.toml b/pyproject.tomlindex 420661e2..60f30233 100644--- a/pyproject.toml+++ b/pyproject.toml@@ -54,7 +54,21 @@ structured = ["pydantic>=2.0.0"]# `pip install laya[onnx]` adds ONNX Runtime for `laya.onnx_agent` and `onnxscript`, which# `torch.onnx.export` requires (`scripts/export_onnx.py`, `tests/test_onnx.py`).-onnx = ["onnx", "onnxruntime", "onnxscript"]+#+# The NumPy cap is not about ONNX: the extra's own resolution (onnx -> ml-dtypes) otherwise+# pulls NumPy 2, and torch below 2.3 is built against NumPy 1.x and raises "Numpy is not+# available" when it converts a tensor against 2.x, which `predict()` needs for its final+# `.numpy()`. This package declares `torch>=2.0.0`, so the extra must not force a NumPy the+# declared floor cannot use.+#+# It is scoped by marker rather than written as a blanket `numpy<2`, because a hard upper cap+# binds every `laya[onnx]` install -- including the many on a current torch -- and makes the+# extra unresolvable next to anything else requiring `numpy>=2`. The marker names the stack this+# actually affects: the macOS/Intel checkout (#948), where torch tops out at 2.2.2 and NumPy 1.26+# satisfies onnx, onnxruntime and onnxscript. Every other platform resolves what `onnx` wants,+# so none of those three is quietly pinned to NumPy 1.x on its behalf.+onnx = ["onnx", "onnxruntime", "onnxscript",+ "numpy<2; platform_system == 'Darwin' and platform_machine == 'x86_64'"]# `pip install laya[langchain]` adds the LangChain / LangGraph integrations (laya.integrations).langchain = [