Merge branch 'main' into jz/cache-log

Author: jackzhxng

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-27e35de6c288bffad1b4d18b393579c1d1a95547
+08434df1f2f88c9770e59246caa2ff9c6f613270

.ci/docker/common/install_java.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -ex
+
+apt-get update
+
+apt-get install -y --no-install-recommends openjdk-17-jdk

.ci/docker/ubuntu/Dockerfile

@@ -30,6 +30,10 @@ ARG BUCK2_VERSION
 COPY ./common/install_buck.sh install_buck.sh
 RUN bash ./install_buck.sh && rm install_buck.sh
 
+# Install java
+COPY ./common/install_java.sh install_java.sh
+RUN bash ./install_java.sh && rm install_java.sh
+
 # Setup user
 COPY ./common/install_user.sh install_user.sh
 RUN bash ./install_user.sh && rm install_user.sh

.ci/scripts/build_llama_android.sh

@@ -42,6 +42,10 @@ install_executorch_and_backend_lib() {
 
 build_llama_runner() {
     echo "Building llama runner for Android..."
+    pushd extension/llm/tokenizers
+    echo "Updating tokenizers submodule"
+    git submodule update --init
+    popd
     ANDROID_ABI=arm64-v8a
     cmake -DBUCK2="${BUCK2}" \
     -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK"/build/cmake/android.toolchain.cmake  \

.ci/scripts/gather_test_models.py

@@ -30,6 +30,7 @@
         "dl3": "linux.4xlarge.memory",
         "emformer_join": "linux.4xlarge.memory",
         "emformer_predict": "linux.4xlarge.memory",
+        "phi-4-mini": "linux.4xlarge.memory",
     }
 }
 

.ci/scripts/test_ane_static_llama.sh

@@ -0,0 +1,33 @@
+#!/bin/bash
+# Copyright (c) Qualcomm Innovation Center, Inc.
+# All rights reserved
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -exu
+
+source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
+
+export EXECUTORCH_ROOT="$(dirname "${BASH_SOURCE[0]}")/../.."
+
+if [[ -z "${PYTHON_EXECUTABLE:-}" ]]; then
+  PYTHON_EXECUTABLE=python3
+fi
+
+which "${PYTHON_EXECUTABLE}"
+
+# Update tokenizers submodule
+pushd $EXECUTORCH_ROOT/extension/llm/tokenizers
+echo "Update tokenizers submodule"
+git submodule update --init
+popd
+
+pushd $EXECUTORCH_ROOT/examples/apple/coreml/llama
+
+# Download stories llama110m artifacts
+download_stories_model_artifacts
+
+python export.py -n model.pte -p params.json -c stories110M.pt --seq_length 32 --max_seq_length 64 --dtype fp16 --coreml-quantize c4w
+
+popd

.ci/scripts/test_llama.sh

@@ -173,6 +173,10 @@ cmake_install_executorch_libraries() {
 
 cmake_build_llama_runner() {
     echo "Building llama runner"
+    pushd extension/llm/tokenizers
+    echo "Updating tokenizers submodule"
+    git submodule update --init
+    popd
     dir="examples/models/llama"
     retry cmake \
         -DCMAKE_INSTALL_PREFIX=cmake-out \

.ci/scripts/test_model.sh

@@ -100,6 +100,15 @@ test_model() {
       rm "./${MODEL_NAME}.pte"
       return  # Skip running with portable executor runnner since portable doesn't support Qwen's biased linears.
   fi
+  if [[ "${MODEL_NAME}" == "phi-4-mini" ]]; then
+      # Install requirements for export_llama
+      bash examples/models/llama/install_requirements.sh
+      # Test export_llama script: python3 -m examples.models.llama.export_llama.
+      "${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama --model "${MODEL_NAME}" -c examples/models/llama/params/demo_rand_params.pth -p examples/models/phi-4-mini/config.json
+      run_portable_executor_runner
+      rm "./${MODEL_NAME}.pte"
+      return
+  fi
 
   # Export a basic .pte and run the model.
   "${PYTHON_EXECUTABLE}" -m examples.portable.scripts.export --model_name="${MODEL_NAME}" "${STRICT}"
@@ -164,6 +173,7 @@ test_model_with_qnn() {
   export LD_LIBRARY_PATH=$QNN_SDK_ROOT/lib/x86_64-linux-clang/
   export PYTHONPATH=$EXECUTORCH_ROOT/..
 
+  EXTRA_FLAGS=""
   if [[ "${MODEL_NAME}" == "dl3" ]]; then
     EXPORT_SCRIPT=deeplab_v3
   elif [[ "${MODEL_NAME}" == "mv3" ]]; then
@@ -176,6 +186,12 @@ test_model_with_qnn() {
     EXPORT_SCRIPT=inception_v3
   elif [[ "${MODEL_NAME}" == "vit" ]]; then
     EXPORT_SCRIPT=torchvision_vit
+  elif [[ "${MODEL_NAME}" == "mb" ]]; then
+    EXPORT_SCRIPT=mobilebert_fine_tune
+    EXTRA_FLAGS="--num_epochs 1"
+    pip install scikit-learn
+  elif [[ "${MODEL_NAME}" == "w2l" ]]; then
+    EXPORT_SCRIPT=wav2letter
   elif [[ "${MODEL_NAME}" == "edsr" ]]; then
     EXPORT_SCRIPT=edsr
     # Additional deps for edsr
@@ -189,7 +205,7 @@ test_model_with_qnn() {
   # 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 ${QNN_CHIPSET} --compile_only
+  "${PYTHON_EXECUTABLE}" -m examples.qualcomm.scripts.${EXPORT_SCRIPT} -b ${CMAKE_OUTPUT_DIR} -m ${QNN_CHIPSET} --compile_only $EXTRA_FLAGS
   EXPORTED_MODEL=$(find "./${EXPORT_SCRIPT}" -type f -name "${MODEL_NAME}*.pte" -print -quit)
 }
 

.ci/scripts/unittest-buck2.sh

@@ -14,5 +14,7 @@ buck2 query "//backends/apple/... + //backends/example/... + \
 //runtime/... + //schema/... + //test/... + //util/..."
 
 # TODO: expand the covered scope of Buck targets.
-buck2 build //runtime/core/portable_type/...
-buck2 test //runtime/core/portable_type/...
+# //runtime/kernel/... is failing because //third-party:torchgen_files's shell script can't find python on PATH.
+# //runtime/test/... requires Python torch, which we don't have in our OSS buck setup.
+buck2 build //runtime/backend/... //runtime/core/... //runtime/executor: //runtime/kernel/... //runtime/platform/...
+buck2 test //runtime/backend/... //runtime/core/... //runtime/executor: //runtime/kernel/... //runtime/platform/...

.github/workflows/android-perf.yml

@@ -96,6 +96,63 @@ jobs:
 
           PYTHONPATH="${PWD}" python .ci/scripts/gather_benchmark_configs.py $ARGS
 
+  prepare-test-specs:
+    runs-on: linux.2xlarge
+    needs: set-parameters
+    strategy:
+      matrix: ${{ fromJson(needs.set-parameters.outputs.benchmark_configs) }}
+      fail-fast: false
+    steps:
+      - uses: actions/checkout@v3
+
+      - name: Prepare the spec
+        id: prepare
+        shell: bash
+        env:
+          BENCHMARK_CONFIG: ${{ toJSON(matrix) }}
+        working-directory: extension/benchmark/android/benchmark
+        run: |
+          set -eux
+
+          # The model will be exported in the next step to this S3 path
+          MODEL_PATH="https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/${{ matrix.model }}_${{ matrix.config }}/model.zip"
+          # We could write a script to properly use jinja here, but there is only one variable,
+          # so let's just sed it
+          sed -i -e 's,{{ model_path }},'"${MODEL_PATH}"',g' android-llm-device-farm-test-spec.yml.j2
+
+          BENCHMARK_CONFIG_ID=$(echo "${{ matrix.model }}_${{ matrix.config }}" | sed -e 's/[^A-Za-z0-9._-]/_/g')
+          # The config for this benchmark runs, we save it in the test spec so that it can be fetched
+          # later by the upload script
+          sed -i -e 's,{{ benchmark_config_id }},'"${BENCHMARK_CONFIG_ID}"',g' android-llm-device-farm-test-spec.yml.j2
+
+          cp android-llm-device-farm-test-spec.yml.j2 android-llm-device-farm-test-spec.yml
+          # Just print the test spec for debugging
+          cat android-llm-device-farm-test-spec.yml
+
+          # Save the benchmark configs so that we can use it later in the dashboard
+          echo "${BENCHMARK_CONFIG}" > "${BENCHMARK_CONFIG_ID}.json"
+          echo "benchmark-config-id=${BENCHMARK_CONFIG_ID}" >> $GITHUB_OUTPUT
+
+      - name: Upload the spec
+        uses: seemethere/upload-artifact-s3@v5
+        with:
+          s3-bucket: gha-artifacts
+          s3-prefix: |
+            ${{ github.repository }}/${{ github.run_id }}/artifacts/${{ matrix.model }}_${{ matrix.config }}
+          retention-days: 1
+          if-no-files-found: error
+          path: extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml
+
+      - name: Update the benchmark configs
+        uses: seemethere/upload-artifact-s3@v5
+        with:
+          s3-bucket: gha-artifacts
+          s3-prefix: |
+            ${{ github.repository }}/${{ github.run_id }}/artifacts/benchmark-configs/
+          retention-days: 1
+          if-no-files-found: error
+          path: extension/benchmark/android/benchmark/${{ steps.prepare.outputs.benchmark-config-id }}.json
+
   export-models:
     name: export-models
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
@@ -278,69 +335,6 @@ jobs:
         fi
         echo "::endgroup::"
 
-  prepare-test-specs:
-    runs-on: linux.2xlarge
-    needs:
-      - set-parameters
-      - export-models
-    strategy:
-      matrix: ${{ fromJson(needs.set-parameters.outputs.benchmark_configs) }}
-      fail-fast: false
-    steps:
-      - uses: actions/checkout@v3
-
-      - name: Prepare the spec
-        id: prepare
-        shell: bash
-        env:
-          BENCHMARK_CONFIG: ${{ toJSON(matrix) }}
-        working-directory: extension/benchmark/android/benchmark
-        run: |
-          set -eux
-
-          # The model will be exported in the next step to this S3 path
-          MODEL_PATH="https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/${{ matrix.model }}_${{ matrix.config }}/model.zip"
-
-          # Check if the model artifact exists, fail this step skip generating test-spec.
-          curl -s --head -f ${MODEL_PATH}
-
-          # We could write a script to properly use jinja here, but there is only one variable,
-          # so let's just sed it
-          sed -i -e 's,{{ model_path }},'"${MODEL_PATH}"',g' android-llm-device-farm-test-spec.yml.j2
-
-          BENCHMARK_CONFIG_ID=$(echo "${{ matrix.model }}_${{ matrix.config }}" | sed -e 's/[^A-Za-z0-9._-]/_/g')
-          # The config for this benchmark runs, we save it in the test spec so that it can be fetched
-          # later by the upload script
-          sed -i -e 's,{{ benchmark_config_id }},'"${BENCHMARK_CONFIG_ID}"',g' android-llm-device-farm-test-spec.yml.j2
-
-          cp android-llm-device-farm-test-spec.yml.j2 android-llm-device-farm-test-spec.yml
-          # Just print the test spec for debugging
-          cat android-llm-device-farm-test-spec.yml
-
-          # Save the benchmark configs so that we can use it later in the dashboard
-          echo "${BENCHMARK_CONFIG}" > "${BENCHMARK_CONFIG_ID}.json"
-          echo "benchmark-config-id=${BENCHMARK_CONFIG_ID}" >> $GITHUB_OUTPUT
-
-      - name: Upload the spec
-        uses: seemethere/upload-artifact-s3@v5
-        with:
-          s3-bucket: gha-artifacts
-          s3-prefix: |
-            ${{ github.repository }}/${{ github.run_id }}/artifacts/${{ matrix.model }}_${{ matrix.config }}
-          retention-days: 1
-          if-no-files-found: error
-          path: extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml
-
-      - name: Update the benchmark configs
-        uses: seemethere/upload-artifact-s3@v5
-        with:
-          s3-bucket: gha-artifacts
-          s3-prefix: |
-            ${{ github.repository }}/${{ github.run_id }}/artifacts/benchmark-configs/
-          retention-days: 1
-          if-no-files-found: error
-          path: extension/benchmark/android/benchmark/${{ steps.prepare.outputs.benchmark-config-id }}.json
-
   build-benchmark-app:
     name: build-benchmark-app
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main