Add preprocess to ci

ghstack-source-id: 644f38e
Pull Request resolved: #6544

Author: lucylq

.github/workflows/_unittest.yml

@@ -37,6 +37,9 @@ jobs:
         CMAKE_ARGS="-DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
         .ci/scripts/setup-linux.sh cmake
 
+        # Install llama3_2_vision dependencies.
+        bash examples/models/llama3_2_vision/install_requirements.sh
+
         # Run pytest with coverage
         pytest -n auto --cov=./ --cov-report=xml
         # Run gtest
@@ -67,6 +70,9 @@ jobs:
         ${CONDA_RUN} --no-capture-output \
         .ci/scripts/setup-macos.sh cmake
 
+        # Install llama3_2_vision dependencies.
+        bash examples/models/llama3_2_vision/install_requirements.sh
+
         # Run pytest with coverage
         ${CONDA_RUN} pytest -n auto --cov=./ --cov-report=xml
         # Run gtest

examples/models/llama3_2_vision/install_requirements.sh

@@ -10,4 +10,4 @@ pip install --pre torchtune --extra-index-url https://download.pytorch.org/whl/n
 
 # Install torchao.
 TORCHAO_VERSION=$(cat "$(dirname "$0")"/../../../.ci/docker/ci_commit_pins/torchao.txt)
-pip install --no-use-pep517 "git+https://github.com/pytorch/ao.git@${TORCHAO_VERSION}"
+pip install "git+https://github.com/pytorch/ao.git@${TORCHAO_VERSION}"

examples/models/llama3_2_vision/preprocess/model.py

@@ -26,6 +26,9 @@ class PreprocessConfig:
     max_num_tiles: int = 4
     tile_size: int = 224
     antialias: bool = False
+    # Used for reference eager model from torchtune.
+    resize_to_max_canvas: bool = False
+    possible_resolutions: Optional[List[Tuple[int, int]]] = None
 
 
 class CLIPImageTransformModel(EagerModelBase):