Merge branch 'main' into upload-ios-perf-results

Author: huydhn

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-4b2970f7cd3cdd56883cacf116a8693862f89db5
+d1b87e26e5c4343f5b56bb1e6f89b479b389bfac

.ci/docker/requirements-ci.txt

@@ -1,5 +1,5 @@
 mpmath==1.3.0
-numpy==1.21.3; python_version == '3.10'
+numpy==1.22.0; python_version == '3.10'
 numpy==1.23.2; python_version == '3.11'
 numpy; python_version >= '3.12'
 PyYAML==6.0.1

.ci/scripts/test_llama.sh

@@ -171,7 +171,7 @@ else
 fi
 
 # Check dtype.
-EXPORTED_MODEL_NAME="llama2"
+EXPORTED_MODEL_NAME="tinyllama_${MODE}_${DTYPE}"
 if [[ "${DTYPE}" == "fp16" ]]; then
   EXPORTED_MODEL_NAME="${EXPORTED_MODEL_NAME}_h"
 elif [[ "${DTYPE}" == "bf16" ]]; then

.ci/scripts/test_model.sh

@@ -155,30 +155,24 @@ test_model_with_qnn() {
 
   if [[ "${MODEL_NAME}" == "dl3" ]]; then
     EXPORT_SCRIPT=deeplab_v3
-    EXPORTED_MODEL_NAME=dlv3_qnn.pte
   elif [[ "${MODEL_NAME}" == "mv3" ]]; then
     EXPORT_SCRIPT=mobilenet_v3
-    EXPORTED_MODEL_NAME=mv3_qnn.pte
   elif [[ "${MODEL_NAME}" == "mv2" ]]; then
     EXPORT_SCRIPT=mobilenet_v2
-    EXPORTED_MODEL_NAME=mv2_qnn.pte
   elif [[ "${MODEL_NAME}" == "ic4" ]]; then
     EXPORT_SCRIPT=inception_v4
-    EXPORTED_MODEL_NAME=ic4_qnn.pte
   elif [[ "${MODEL_NAME}" == "ic3" ]]; then
     EXPORT_SCRIPT=inception_v3
-    EXPORTED_MODEL_NAME=ic3_qnn.pte
   elif [[ "${MODEL_NAME}" == "vit" ]]; then
     EXPORT_SCRIPT=torchvision_vit
-    EXPORTED_MODEL_NAME=vit_qnn.pte
   fi
 
   # Use SM8450 for S22, SM8550 for S23, and SM8560 for S24
   # TODO(guangyang): Make QNN chipset matches the target device
   QNN_CHIPSET=SM8450
 
   "${PYTHON_EXECUTABLE}" -m examples.qualcomm.scripts.${EXPORT_SCRIPT} -b ${CMAKE_OUTPUT_DIR} -m ${QNN_CHIPSET} --compile_only
-  EXPORTED_MODEL=./${EXPORT_SCRIPT}/${EXPORTED_MODEL_NAME}
+  EXPORTED_MODEL=$(find "./${EXPORT_SCRIPT}" -type f -name "${MODEL_NAME}*.pte" -print -quit)
 }
 
 test_model_with_coreml() {
@@ -187,7 +181,24 @@ test_model_with_coreml() {
     exit 1
   fi
 
-  "${PYTHON_EXECUTABLE}" -m examples.apple.coreml.scripts.export --model_name="${MODEL_NAME}"
+  DTYPE=float16
+
+  "${PYTHON_EXECUTABLE}" -m examples.apple.coreml.scripts.export --model_name="${MODEL_NAME}" --compute_precision "${DTYPE}"
+  EXPORTED_MODEL=$(find "." -type f -name "${MODEL_NAME}*.pte" -print -quit)
+  # TODO:
+  if [ -n "$EXPORTED_MODEL" ]; then
+    EXPORTED_MODEL_WITH_DTYPE="${EXPORTED_MODEL%.pte}_${DTYPE}.pte"
+    mv "$EXPORTED_MODEL" "$EXPORTED_MODEL_WITH_DTYPE"
+    EXPORTED_MODEL="$EXPORTED_MODEL_WITH_DTYPE"
+    echo "Renamed file path: $EXPORTED_MODEL"
+  else
+    echo "No .pte file found"
+    exit 1
+  fi
+}
+
+test_model_with_mps() {
+  "${PYTHON_EXECUTABLE}" -m examples.apple.mps.scripts.mps_example --model_name="${MODEL_NAME}" --use_fp16
   EXPORTED_MODEL=$(find "." -type f -name "${MODEL_NAME}*.pte" -print -quit)
 }
 
@@ -206,6 +217,12 @@ elif [[ "${BACKEND}" == "coreml" ]]; then
   if [[ $? -eq 0 ]]; then
     prepare_artifacts_upload
   fi
+elif [[ "${BACKEND}" == "mps" ]]; then
+  echo "Testing ${MODEL_NAME} with mps..."
+  test_model_with_mps
+  if [[ $? -eq 0 ]]; then
+    prepare_artifacts_upload
+  fi
 elif [[ "${BACKEND}" == "xnnpack" ]]; then
   echo "Testing ${MODEL_NAME} with xnnpack..."
   WITH_QUANTIZATION=true

.github/workflows/android-perf.yml

@@ -292,7 +292,7 @@ jobs:
               --output-dir benchmark-results \
               --repo ${{ github.repository }} \
               --head-branch ${{ github.head_ref || github.ref_name }} \
-              --workflow-name ${{ github.workflow }} \
+              --workflow-name "${{ github.workflow }}" \
               --workflow-run-id ${{ github.run_id }} \
               --workflow-run-attempt ${{ github.run_attempt }}
           done

.github/workflows/android.yml

@@ -15,6 +15,7 @@ on:
       - install_requirements.sh
       - examples/demo-apps/android/**
       - extension/android/**
+      - extension/benchmark/android/**
       - extension/module/**
   workflow_dispatch:
 

.github/workflows/apple-perf.yml

@@ -169,6 +169,8 @@ jobs:
             DELEGATE_CONFIG="xnnpack+custom+qe"
           elif [[ ${{ matrix.delegate }} == "coreml" ]]; then
             DELEGATE_CONFIG="coreml"
+          elif [[ ${{ matrix.delegate }} == "mps" ]]; then
+            DELEGATE_CONFIG="mps"
           fi
           PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
             bash .ci/scripts/test_llama.sh "${{ matrix.model }}" "${BUILD_MODE}" "${DTYPE}" "${DELEGATE_CONFIG}" "${ARTIFACTS_DIR_NAME}"
@@ -235,17 +237,17 @@ jobs:
         PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
           build/build_apple_frameworks.sh --coreml --custom --mps --optimized --portable --quantized --xnnpack
 
-        mkdir -p extension/apple/Benchmark/Frameworks
+        mkdir -p extension/benchmark/apple/Benchmark/Frameworks
         for FRAMEWORK in "${FRAMEWORKS[@]}"; do (
-          cp -r "cmake-out/${FRAMEWORK}.xcframework" extension/apple/Benchmark/Frameworks/
+          cp -r "cmake-out/${FRAMEWORK}.xcframework" extension/benchmark/apple/Benchmark/Frameworks/
         ) done
         echo "::endgroup::"
 
         # NB: Although exported models can be copied to this directory and bundled together with the
         # app, we don't use this in CI and rely on AWS extra data parameter to make the model and the
         # tokenizer available to the benchmark. This decouples the app and the model. We just need to
         # create the directory here to pass the build
-        mkdir -p extension/apple/Benchmark/Models
+        mkdir -p extension/benchmark/apple/Benchmark/Models
         ${CONDA_RUN} --no-capture-output \
           build/build_apple_llm_demo.sh ${ARTIFACTS_DIR_NAME}
 

.github/workflows/apple.yml

@@ -18,6 +18,7 @@ on:
       - build/test_ios_ci.sh
       - examples/demo-apps/apple_ios/**
       - extension/apple/**
+      - extension/benchmark/apple/**
       - extension/module/**
   workflow_dispatch:
 
@@ -272,14 +273,14 @@ jobs:
         PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
           build/build_apple_frameworks.sh --coreml --custom --mps --optimized --portable --quantized --xnnpack
 
-        mkdir -p extension/apple/Benchmark/Frameworks
+        mkdir -p extension/benchmark/apple/Benchmark/Frameworks
         for FRAMEWORK in "${FRAMEWORKS[@]}"; do (
-          cp -r "cmake-out/${FRAMEWORK}.xcframework" extension/apple/Benchmark/Frameworks/
+          cp -r "cmake-out/${FRAMEWORK}.xcframework" extension/benchmark/apple/Benchmark/Frameworks/
         ) done
         echo "::endgroup::"
 
         echo "::group::Build ExecuTorch benchmark app"
-        mkdir -p extension/apple/Benchmark/Models
+        mkdir -p extension/benchmark/apple/Benchmark/Models
         ${CONDA_RUN} --no-capture-output \
           build/build_apple_llm_demo.sh ${ARTIFACTS_DIR_NAME}
         echo "::endgroup::"

.github/workflows/lint.yml

@@ -66,7 +66,7 @@ jobs:
         FILES_NEEDS_FORMAT=$(/opt/google-java-format -n extension/android/src/main/java/org/pytorch/executorch/*.java \
           examples/demo-apps/android/ExecuTorchDemo/app/src/main/java/com/example/executorchdemo/*.java \
           examples/demo-apps/android/LlamaDemo/app/src/main/java/com/example/executorchllamademo/*.java \
-          extension/android/benchmark/app/src/main/java/org/pytorch/minibench/*.java)
+          extension/benchmark/android/benchmark/app/src/main/java/org/pytorch/minibench/*.java)
         if [ -n "$FILES_NEEDS_FORMAT" ]; then
           echo "Warning: The following files need formatting. Please use google-java-format."
           echo "Use a binary from https://github.com/google/google-java-format/releases/"

.github/workflows/trunk.yml

@@ -311,8 +311,8 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
         PYTHON_EXECUTABLE=python bash .ci/scripts/test_model.sh ${{ matrix.model }} "cmake" "qnn"
 
-  test-coreml-model:
-    name: test-coreml-model
+  test-apple-model:
+    name: test-apple-model
     uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
     strategy:
       fail-fast: false
@@ -324,20 +324,25 @@ jobs:
       timeout: 90
       script: |
         BUILD_TOOL=cmake
-        BACKEND=coreml
 
         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}"
         PYTHON_EXECUTABLE=python ${CONDA_RUN} bash backends/apple/coreml/scripts/install_requirements.sh
         echo "Finishing installing coreml."
+        PYTHON_EXECUTABLE=python ${CONDA_RUN} bash backends/apple/mps/install_requirements.sh
+        echo "Finishing installing mps."
 
         # Build and test coreml model
         MODELS=(mv3 ic4 resnet50 edsr mobilebert w2l)
         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}" "${BUILD_TOOL}" "${BACKEND}"
+          PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/test_model.sh "${MODEL_NAME}" "${BUILD_TOOL}" "coreml"
+          echo "::endgroup::"
+
+          echo "::group::Exporting mps model: $MODEL_NAME"
+          PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/test_model.sh "${MODEL_NAME}" "${BUILD_TOOL}" "mps"
           echo "::endgroup::"
         done
 

.github/workflows/upload-android-test-specs.yml

@@ -4,13 +4,13 @@ on:
   pull_request:
     paths:
       - .github/workflows/upload-android-test-specs.yml
-      - extension/android/benchmark/android-llm-device-farm-test-spec.yml
+      - extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml
   push:
     branches:
       - main
     paths:
       - .github/workflows/upload-android-test-specs.yml
-      - extension/android/benchmark/android-llm-device-farm-test-spec.yml
+      - extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml
 
 concurrency:
   # NB: This concurency group needs to be different than the one used in android-perf, otherwise
@@ -32,7 +32,7 @@ jobs:
             ${{ github.repository }}/${{ github.run_id }}/artifacts
           retention-days: 1
           if-no-files-found: error
-          path: extension/android/benchmark/android-llm-device-farm-test-spec.yml
+          path: extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml
 
   validate-android-test-spec:
     needs: upload-android-test-spec-for-validation
@@ -77,7 +77,7 @@ jobs:
 
       - name: Upload the spec to S3 ossci-android bucket
         shell: bash
-        working-directory: extension/android/benchmark/
+        working-directory: extension/benchmark/android/benchmark/
         env:
           SPEC_FILE: android-llm-device-farm-test-spec.yml
         run: |

.github/workflows/upload-apple-test-specs.yml

@@ -4,13 +4,13 @@ on:
   pull_request:
     paths:
       - .github/workflows/upload-apple-test-specs.yml
-      - examples/demo-apps/apple_ios/default-ios-device-farm-appium-test-spec.yml
+      - extension/benchmark/apple/Benchmark/default-ios-device-farm-appium-test-spec.yml
   push:
     branches:
       - main
     paths:
       - .github/workflows/upload-apple-test-specs.yml
-      - examples/demo-apps/apple_ios/default-ios-device-farm-appium-test-spec.yml
+      - extension/benchmark/apple/Benchmark/default-ios-device-farm-appium-test-spec.yml
 
 concurrency:
   # NB: This concurency group needs to be different than the one used in apple-perf, otherwise
@@ -32,7 +32,7 @@ jobs:
             ${{ github.repository }}/${{ github.run_id }}/artifacts
           retention-days: 1
           if-no-files-found: error
-          path: examples/demo-apps/apple_ios/default-ios-device-farm-appium-test-spec.yml
+          path: extension/benchmark/apple/Benchmark/default-ios-device-farm-appium-test-spec.yml
 
   validate-apple-test-spec:
     needs: upload-apple-test-spec-for-validation
@@ -78,7 +78,7 @@ jobs:
 
       - name: Upload the spec to S3 ossci-ios bucket
         shell: bash
-        working-directory: examples/demo-apps/apple_ios
+        working-directory: extension/benchmark/apple/Benchmark/
         env:
           SPEC_FILE: default-ios-device-farm-appium-test-spec.yml
         run: |

.lintrunner.toml

@@ -180,6 +180,7 @@ exclude_patterns = [
     '**/*.bat',
     '**/*.jpg',
     '**/*.jar',
+    '**/*.gif',
     # File contains @generated
     'extension/llm/custom_ops/spinquant/fast_hadamard_transform_special.h',
     'extension/llm/custom_ops/spinquant/test/fast_hadamard_transform_special_unstrided_cpu.h',