Add more models to coreml testing

Author: mergennachin

.ci/scripts/test_model.sh

@@ -91,10 +91,11 @@ test_model() {
     run_portable_executor_runner
     rm "./${MODEL_NAME}.pte"
   fi
-  STRICT="--no-strict"
+  STRICT="--strict"
   if [[ "${MODEL_NAME}" == "llava" ]]; then
     # Install requirements for llava
     bash examples/models/llava/install_requirements.sh
+    STRICT="--no-strict"
   fi
   if [[ "${MODEL_NAME}" == "qwen2_5" ]]; then
       # Install requirements for export_llama

.github/workflows/trunk.yml

@@ -18,8 +18,8 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  test-models-macos:
-    name: test-models-macos
+  test-models-macos-cpu:
+    name: test-models-macos-cpu
     uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
     strategy:
       matrix:
@@ -55,6 +55,31 @@ jobs:
         # Build and test executorch
         PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/test_model.sh "${MODEL_NAME}" "${BUILD_TOOL}" "${BACKEND}"
 
+  test-models-macos-coreml:
+    name: test-models-macos-coreml
+    uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
+    strategy:
+      matrix: 
+        model: [dl3, ic3, mv3, resnet50, vit, w2l, edsr, mobilebert, emformer_join, emformer_transcribe]
+        backend: [coreml]
+      fail-fast: false
+    with:
+      runner: macos-m1-stable
+      python-version: '3.11'
+      submodules: 'recursive'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 90
+      script: |
+        MODEL_NAME=${{ matrix.model }}
+        BUILD_TOOL=cmake
+        BACKEND=${{ matrix.backend }}
+
+        bash .ci/scripts/setup-conda.sh
+        # Setup MacOS dependencies as there is no Docker support on MacOS atm
+        PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/setup-macos.sh --build-tool "${BUILD_TOOL}"
+        # Build and test executorch
+        PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/test_model.sh "${MODEL_NAME}" "${BUILD_TOOL}" "${BACKEND}"
+
   test-models-linux-aarch64:
     name: test-models-linux-aarch64
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
@@ -589,7 +614,7 @@ jobs:
         echo "Finishing installing coreml."
 
         # Build and test coreml model
-        MODELS=(mv3 ic4 resnet50 edsr mobilebert w2l)
+        MODELS=(mv3 ic4 resnet50)
         for MODEL_NAME in "${MODELS[@]}"; do
           echo "::group::Exporting coreml model: $MODEL_NAME"
           PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/test_model.sh "${MODEL_NAME}" "${BUILD_TOOL}" "coreml"

examples/xnnpack/aot_compiler.py

@@ -87,14 +87,14 @@
 
     model = model.eval()
     # pre-autograd export. eventually this will become torch.export
-    ep = torch.export.export_for_training(model, example_inputs, strict=False)
+    ep = torch.export.export_for_training(model, example_inputs, strict=True)
     model = ep.module()
 
     if args.quantize:
         logging.info("Quantizing Model...")
         # TODO(T165162973): This pass shall eventually be folded into quantizer
         model = quantize(model, example_inputs, quant_type)
-        ep = torch.export.export_for_training(model, example_inputs, strict=False)
+        ep = torch.export.export_for_training(model, example_inputs, strict=True)
 
     edge = to_edge_transform_and_lower(
         ep,