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
feat(ci): run the test suite on Windows as well as Linux (#212)
PerryLink4931d7977e0f
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.ymlindex 9a06e70e..c15af608 100644--- a/.github/workflows/ci.yml+++ b/.github/workflows/ci.yml@@ -74,6 +74,67 @@ jobs:- name: Email cleaning testsrun: python tests/test_email.py+ # 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).+ #+ # A separate job rather than an extra axis on `test`, deliberately: it leaves that job's name,+ # and therefore the `tests (pyX.Y)` required checks, exactly as they are. It runs one Python+ # because the variable that was untested is the OS, not the interpreter.+ test-windows:+ name: tests (windows)+ runs-on: windows-latest+ steps:+ - uses: actions/checkout@v4+ - uses: actions/setup-python@v5+ with:+ python-version: "3.11"+ cache: pip++ # 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.+ - name: Install+ run: |+ python -m pip install --upgrade pip+ pip install torch+ pip install -e .+ shell: bash++ # bash on this runner so the quoting rules match the Linux job; the default shell on a+ # Windows runner is PowerShell, where `python -c "...; ..."` is not the same command.+ - name: Import check+ run: python -c "import laya; print(laya.__version__); print(sorted(laya.__all__))"+ shell: bash++ # Same suites as the Linux job. Guarded with -f so a suite added by another open branch+ # cannot make this job red on a tree that does not have it yet.+ #+ # test_docker_entrypoint is skipped here on purpose. It drives docker/entrypoint.py, which+ # only ever runs inside a Linux container, through `os.execvp`. On Windows exec* rebuilds+ # the command line from argv without re-applying the quoting `subprocess` used to build it,+ # so its `-c "<script>"` argument arrives split at the first space and the child dies with+ # "SyntaxError: invalid syntax" before printing. The environment passthrough the suite+ # exists to check is unaffected, and switching the entrypoint to `subprocess.run` makes it+ # pass -- a change to a container script, so it belongs in its own PR, not this one.+ - name: Routing and language-detection tests+ run: |+ for t in test_router test_criteria test_download test_shortlist \+ test_decision_model test_packaging test_lang_guess \+ test_calibration_persistence test_context_manager \+ test_criteria_normalization test_docker_entrypoint \+ test_empty_questions test_router_memory test_shortlist_cosine \+ test_temperature_loading; do+ if [ "$t" = "test_docker_entrypoint" ]; then+ echo "skipping $t: docker/entrypoint.py is Linux-container only"+ continue+ fi+ if [ -f "tests/$t.py" ]; then python "tests/$t.py"; fi+ done+ shell: bash++ - name: Email cleaning tests+ run: python tests/test_email.py+ shell: bash+lint:name: lintruns-on: ubuntu-latest