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
ci: run the pytest suites, which no lane was executing
PerryLinkbc2a12982740
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.ymlindex ce48ff4f..8e019ada 100644--- a/.github/workflows/ci.yml+++ b/.github/workflows/ci.yml@@ -71,8 +71,11 @@ jobs:python -m pip install --upgrade pip# CPU-only torch keeps the job under a minute instead of pulling CUDA wheelspip install torch --index-url https://download.pytorch.org/whl/cpu- # Install with the mcp extra so laya[mcp] resolution and the laya.mcp layer are both exercised- pip install -e ".[mcp]"+ # Install with the mcp extra so laya[mcp] resolution and the laya.mcp layer are both exercised,+ # and the serve extra so the HTTP surface is installed rather than skipped.+ pip install -e ".[mcp,serve]"+ # The pytest-style suites need the runner; nothing else in this job uses it.+ pip install pytest- name: Import checkrun: python -c "import laya; print(laya.__version__); print(sorted(laya.__all__))"@@ -114,6 +117,23 @@ jobs:- name: Email cleaning testsrun: python tests/test_email.py+ # These four are pytest suites, and `python tests/<name>.py` does not run them: it defines+ # the test functions and exits 0, having asserted nothing. They were counted by the suite+ # list above without ever executing, which is how the HTTP surface came to have no coverage+ # in CI at all (#374).+ #+ # test_onnx.py and test_fast.py are deliberately not here: both are skip-guarded on an extra+ # this job does not install, so they would pass trivially, and their real execution belongs+ # in a lane that installs `onnx` or has a GPU.+ - name: Pytest suites (serve, batch, system-one, audit regressions, truncation)+ run: |+ python -m pytest tests/test_serve.py \+ tests/test_router_batch.py \+ tests/test_system_one_lang.py \+ tests/test_audit_regressions.py \+ tests/test_truncation_direction.py \+ -q+# Every workflow in this repo ran on ubuntu-latest only, so a test that fails off Linux was# invisible until a user hit it -- which is what happened to tests/test_download.py (#140).#@@ -132,11 +152,14 @@ jobs:# The +cpu index used on Linux is Linux/macOS only. On Windows the default PyPI wheel is# already CPU-only, so installing torch plainly is both correct and what a user gets.+ # The serve extra and pytest are here for the same reason as the Linux job: the HTTP+ # surface is otherwise skipped and the pytest suites otherwise do not run at all.- name: Installrun: |python -m pip install --upgrade pippip install torch- pip install -e .+ pip install -e ".[serve]"+ pip install pytestshell: bash# bash on this runner so the quoting rules match the Linux job; the default shell on a@@ -177,6 +200,19 @@ jobs:python tests/test_hooks_api.pyshell: bash+ # Same five pytest suites as the Linux job. They are named explicitly rather than added to+ # the loop above, because the loop runs each file as a script and that is exactly what does+ # not execute a pytest suite (#374).+ - name: Pytest suites (serve, batch, system-one, audit regressions, truncation)+ run: |+ python -m pytest tests/test_serve.py \+ tests/test_router_batch.py \+ tests/test_system_one_lang.py \+ tests/test_audit_regressions.py \+ tests/test_truncation_direction.py \+ -q+ shell: bash+- name: Email cleaning testsrun: python tests/test_email.pyshell: bash