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 new benchmark audit and tests
GaotianJin681f1fd47dca
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.ymlindex ce48ff4f..b3fee5e1 100644--- a/.github/workflows/ci.yml+++ b/.github/workflows/ci.yml@@ -50,6 +50,9 @@ jobs:python-version: "3.11"- run: python research/benchmarks/feishu_zh/audit.py- run: python -m unittest discover -s research/benchmarks/feishu_zh/tests -v+ # This job installs nothing, so running the audit here is what proves it needs+ # nothing but the standard library. Its test suite imports numpy and runs in `test`.+ - run: python research/benchmarks/zh_short_commands/audit.pytest:name: tests (py${{ matrix.python }})@@ -113,6 +116,11 @@ jobs:- name: Email cleaning testsrun: python tests/test_email.py+ # Full dependency tree here, so the stub pipeline and the metric pins that need numpy+ # run instead of skipping. The audit itself runs in `feishu-benchmark`, where nothing+ # is installed.+ - name: Chinese short-command benchmark tests+ run: python -m unittest discover -s research/benchmarks/zh_short_commands/tests -v# 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).