Test on ORT 1.17 and 1.18

.github/workflows/array-api.yml

@@ -12,6 +12,12 @@ jobs:
     name: Array API test
     timeout-minutes: 90
     runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        environment:
+          - ort117py312
+          - ort118py312
     steps:
       - name: Checkout branch
         uses: actions/checkout@v4
@@ -21,10 +27,12 @@ jobs:
           submodules: recursive
       - name: Set up pixi
         uses: prefix-dev/setup-pixi@ba3bb36eb2066252b2363392b7739741bb777659
+        with:
+          environments: ${{ matrix.environment }}
       - name: Install repository
-        run: pixi run postinstall
+        run: pixi run -e ${{ matrix.environment }} postinstall
       - name: Run Array API tests
-        run: pixi run arrayapitests
+        run: pixi run -e ${{ matrix.environment }} arrayapitests
       - name: Upload Array API tests report
         uses: actions/upload-artifact@v4
         with:

.github/workflows/ci.yml

@@ -34,8 +34,8 @@ jobs:
         environment:
           - py310
           - py311
-          - py312
-          - np1x
+          - ort117py312
+          - ort118py312
     steps:
       - name: Checkout branch
         uses: actions/checkout@v4

docs/intros/modelconversion.rst

@@ -110,3 +110,8 @@ We can use an ONNX backend like onnxruntime to run our model. Here we use onnxru
 
     print(out)
     # array([0, 0, 0, 1, 2, 2, 0, 1, 2, 1, 0, 2, 1, 1, 1, 2, 2, 0, 0, 1, 0, 1, 1, 0, 0, 2, 2, 2, 2, 2])
+
+.. note::
+
+    ONNX backends may not always support the entire ONNX specification and can sometimes miss kernel implementations for specific data types.
+    Since onnxruntime is such a common backend, ndonnx ensures proactively that any model generated using :func:`ndonnx.build` is compatible with at least the two latest minor onnxruntime releases on conda-forge.

ndonnx/_build.py

@@ -36,6 +36,8 @@ def build(
     -------
     out: onnx.ModelProto
         ONNX model
+
+    Note: the generated ONNX model should be runnable on the latest two minor versions of onnxruntime documented in an ndonnx release.
     """
 
     def collect_vars(name: str, arr: _CoreArray | Array):