Create test data for round tripping tests (#18)

Since the repo does not have enough git lfs space to store binary files,
I removed all initializers and use textproto to create the round
tripping tests.

---------

Signed-off-by: Justin Chu <[email protected]>

Author: justinchuby

.github/workflows/main.yml

@@ -51,7 +51,7 @@ jobs:
       - name: Pull Test Data
         run: git lfs pull
       - name: Run tests
-        run: nox -t ${{ matrix.nox-tag }} --forcecolor -- --cov=onnx_ir --cov-report=xml --cov-append --cov-branch -n=auto --junitxml junit.xml
+        run: nox -t ${{ matrix.nox-tag }} --forcecolor -- --cov=src/onnx_ir --cov-report=xml --cov-append --cov-branch -n=auto --junitxml junit.xml
       - name: Upload coverage to Codecov
         if: always()
         uses: codecov/codecov-action@v5

REUSE.toml

@@ -14,6 +14,7 @@ path = [
     "**/*.ipynb",
     "**/*.md",
     "**/*.rst",
+    "**/*.textproto",
     "**/*.toml",
     "**/*.yml",
     "CODEOWNERS",

testdata/e2e_models/README.md

@@ -0,0 +1,5 @@
+# Models for end-to-end testing
+
+The models under this directory are generated with [tools/create_test_model.py](/tools/create_test_model.py). ONNX models have all initializer data stripped and save as the textproto format.
+
+If a particular test requires the initializer data to be present, it should create random weights based on the tensor dtype/shape and add them to the `TensorProto`s in `model.graph.initializer`.