Update on "[Executorch][llama] Change runner to decouple prompt length from sequence

length"


length

Following previous diff now we can utilize entire kv cache to generate more
tokens than max prompt length allowed.

Differential Revision: [D69073908](https://our.internmc.facebook.com/intern/diff/D69073908/)

[ghstack-poisoned]

Author: kimishpatel

.ci/scripts/build_android_instrumentation.sh

@@ -12,30 +12,10 @@ if [[ -z "${PYTHON_EXECUTABLE:-}" ]]; then
 fi
 which "${PYTHON_EXECUTABLE}"
 
-build_android_test() {
-  mkdir -p extension/android/executorch_android/src/androidTest/resources
-  cp extension/module/test/resources/add.pte extension/android/executorch_android/src/androidTest/resources
-  pushd extension/android
-  ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:testDebugUnitTest
-  ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:assembleAndroidTest
-  popd
-}
+mkdir -p "${BUILD_AAR_DIR}"/executorch_android/src/androidTest/resources
+cp extension/module/test/resources/add.pte "${BUILD_AAR_DIR}"/executorch_android/src/androidTest/resources
 
-collect_artifacts_to_be_uploaded() {
-  ARTIFACTS_DIR_NAME="$1"
-  # Collect Java library test
-  JAVA_LIBRARY_TEST_DIR="${ARTIFACTS_DIR_NAME}/library_test_dir"
-  mkdir -p "${JAVA_LIBRARY_TEST_DIR}"
-  cp extension/android/executorch_android/build/outputs/apk/androidTest/debug/*.apk "${JAVA_LIBRARY_TEST_DIR}"
-}
-
-main() {
-  build_android_test
-  if [ -n "$ARTIFACTS_DIR_NAME" ]; then
-    collect_artifacts_to_be_uploaded ${ARTIFACTS_DIR_NAME}
-  fi
-}
-
-if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
-  main "$@"
-fi
+pushd "${BUILD_AAR_DIR}"
+ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:testDebugUnitTest
+ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:assembleAndroidTest
+popd

.ci/scripts/gather_test_models.py

@@ -33,7 +33,7 @@
         "dl3": "linux.4xlarge.memory",
         "emformer_join": "linux.4xlarge.memory",
         "emformer_predict": "linux.4xlarge.memory",
-        "phi-4-mini": "linux.4xlarge.memory",
+        "phi_4_mini": "linux.4xlarge.memory",
     }
 }
 

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -78,7 +78,6 @@ ${PYTHON_EXECUTABLE} -m examples.models.llama.export_llama \
     -qmode "torchao:8da${QLINEAR_BITWIDTH}w" \
     --group_size ${QLINEAR_GROUP_SIZE} \
     -E "torchao:${QEMBEDDING_BITWIDTH},${QEMBEDDING_GROUP_SIZE}" \
-    --disable_dynamic_shape \
     -d fp32
 
 # Test run

.ci/scripts/test_model.sh

@@ -100,11 +100,11 @@ 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
+  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 "${MODEL_NAME}" -c examples/models/llama/params/demo_rand_params.pth -p examples/models/phi-4-mini/config.json
+      "${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

.ci/scripts/unittest-macos-buck2.sh

@@ -19,14 +19,14 @@ export TMP_DIR=$(mktemp -d)
 export PATH="${TMP_DIR}:$PATH"
 trap 'rm -rfv ${TMP_DIR}' EXIT
 
-if [[ "$BUILD_TOOL" == "cmake" ]]; then
-    # Setup MacOS dependencies as there is no Docker support on MacOS atm
-    PYTHON_EXECUTABLE=python \
-    EXECUTORCH_BUILD_PYBIND=ON \
-    CMAKE_ARGS="-DEXECUTORCH_BUILD_COREML=ON -DEXECUTORCH_BUILD_MPS=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
-    ${CONDA_RUN} --no-capture-output \
-    .ci/scripts/setup-macos.sh "$@"
+# Setup MacOS dependencies as there is no Docker support on MacOS atm
+PYTHON_EXECUTABLE=python \
+EXECUTORCH_BUILD_PYBIND=ON \
+CMAKE_ARGS="-DEXECUTORCH_BUILD_COREML=ON -DEXECUTORCH_BUILD_MPS=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+${CONDA_RUN} --no-capture-output \
+.ci/scripts/setup-macos.sh "$@"
 
+if [[ "$BUILD_TOOL" == "cmake" ]]; then
     # Install llama3_2_vision dependencies.
     PYTHON_EXECUTABLE=python \
     ${CONDA_RUN} --no-capture-output \

.ci/scripts/unittest-macos.sh

@@ -60,12 +60,46 @@ install_pytorch_and_domains() {
   # Fetch the target commit
   pushd pytorch || return
   git checkout "${TORCH_VERSION}"
-  git submodule update --init --recursive
 
-  export USE_DISTRIBUTED=1
-  # Then build and install PyTorch
-  python setup.py bdist_wheel
-  pip install "$(echo dist/*.whl)"
+  local system_name=$(uname)
+  if [[ "${system_name}" == "Darwin" ]]; then
+    local platform=$(python -c 'import sysconfig; import platform; v=platform.mac_ver()[0].split(".")[0]; platform=sysconfig.get_platform().split("-"); platform[1]=f"{v}_0"; print("_".join(platform))')
+  fi
+  local python_version=$(python -c 'import platform; v=platform.python_version_tuple(); print(f"{v[0]}{v[1]}")')
+  local torch_release=$(cat version.txt)
+  local torch_short_hash=${TORCH_VERSION:0:7}
+  local torch_wheel_path="cached_artifacts/pytorch/executorch/pytorch_wheels/${system_name}/${python_version}"
+  local torch_wheel_name="torch-${torch_release}%2Bgit${torch_short_hash}-cp${python_version}-cp${python_version}-${platform:-}.whl"
+
+  local cached_torch_wheel="https://gha-artifacts.s3.us-east-1.amazonaws.com/${torch_wheel_path}/${torch_wheel_name}"
+  # Cache PyTorch wheel is only needed on MacOS, Linux CI already has this as part
+  # of the Docker image
+  local torch_wheel_not_found=0
+  if [[ "${system_name}" == "Darwin" ]]; then
+    pip install "${cached_torch_wheel}" || torch_wheel_not_found=1
+  else
+    torch_wheel_not_found=1
+  fi
+
+  # Found no such wheel, we will build it from source then
+  if [[ "${torch_wheel_not_found}" == "1" ]]; then
+    echo "No cached wheel found, continue with building PyTorch at ${TORCH_VERSION}"
+
+    git submodule update --init --recursive
+    USE_DISTRIBUTED=1 python setup.py bdist_wheel
+    pip install "$(echo dist/*.whl)"
+
+    # Only AWS runners have access to S3
+    if command -v aws && [[ -z "${GITHUB_RUNNER:-}" ]]; then
+      for wheel_path in dist/*.whl; do
+        local wheel_name=$(basename "${wheel_path}")
+        echo "Caching ${wheel_name}"
+        aws s3 cp "${wheel_path}" "s3://gha-artifacts/${torch_wheel_path}/${wheel_name}"
+      done
+    fi
+  else
+    echo "Use cached wheel at ${cached_torch_wheel}"
+  fi
 
   # Grab the pinned audio and vision commits from PyTorch
   TORCHAUDIO_VERSION=$(cat .github/ci_commit_pins/audio.txt)

.ci/scripts/utils.sh

@@ -27,16 +27,21 @@ jobs:
         conda activate "${CONDA_ENV}"
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh --build-tool buck2
         export ARTIFACTS_DIR_NAME=artifacts-to-be-uploaded
-
-        mkdir -p ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
-        bash examples/models/llama/install_requirements.sh
-        bash ".ci/scripts/test_llama.sh" -model stories110M -build_tool cmake -dtype fp16 -mode portable -upload ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
+        mkdir -p ${ARTIFACTS_DIR_NAME}/
 
         # Build LLM Demo for Android
         export BUILD_AAR_DIR=aar-out
         mkdir -p $BUILD_AAR_DIR
-        bash build/build_android_library.sh ${ARTIFACTS_DIR_NAME}
-        bash .ci/scripts/build_android_instrumentation.sh ${ARTIFACTS_DIR_NAME}
+        bash scripts/build_android_library.sh
+        cp ${BUILD_AAR_DIR}/executorch.aar $ARTIFACTS_DIR_NAME
+
+        mkdir -p ${ARTIFACTS_DIR_NAME}/library_test_dir
+        bash .ci/scripts/build_android_instrumentation.sh
+        cp ${BUILD_AAR_DIR}/executorch_android/build/outputs/apk/androidTest/debug/executorch_android-debug-androidTest.apk "${ARTIFACTS_DIR_NAME}/library_test_dir"
+
+        mkdir -p ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
+        bash examples/models/llama/install_requirements.sh
+        bash ".ci/scripts/test_llama.sh" -model stories110M -build_tool cmake -dtype fp16 -mode portable -upload ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
 
         mkdir -p examples/demo-apps/android/LlamaDemo/app/libs
         cp aar-out/executorch.aar examples/demo-apps/android/LlamaDemo/app/libs
@@ -96,7 +101,7 @@ jobs:
           curl -O https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/llm_demo/app-debug.apk
           curl -O https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/llm_demo/app-debug-androidTest.apk
           curl -O https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/fp32-xnnpack-custom/model.zip
-          curl -o android-test-debug-androidTest.apk https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/library_test_dir/executorch-debug-androidTest.apk
+          curl -o android-test-debug-androidTest.apk https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/library_test_dir/executorch_android-debug-androidTest.apk
           unzip model.zip
           mv *.pte model.pte
 
@@ -120,7 +125,7 @@ jobs:
         with:
           api-level: ${{ env.API_LEVEL }}
           arch: x86_64
-          script: ./build/run_android_emulator.sh
+          script: ./scripts/run_android_emulator.sh
           # NB: This is to boot the emulator faster following the instructions on
           # https://github.com/ReactiveCircus/android-emulator-runner. The max number
           # of cores we can set is 6, any higher number will be reduced to 6.

.github/workflows/_android.yml

@@ -49,4 +49,6 @@ jobs:
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
       script: |
         set -eux
+        # This is needed to get the prebuilt PyTorch wheel from S3
+        ${CONDA_RUN} --no-capture-output pip install awscli==1.37.21
         .ci/scripts/unittest-macos.sh --build-tool "${{ inputs.build-tool }}" --build-mode "${{ inputs.build-mode }}" --editable "${{ inputs.editable }}"

.github/workflows/_unittest.yml

@@ -363,7 +363,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
 
         mkdir -p aar-out
-        PYTHON_EXECUTABLE=python ANDROID_ABIS="arm64-v8a" BUILD_AAR_DIR=aar-out EXECUTORCH_BUILD_QNN=ON QNN_SDK_ROOT=/tmp/qnn/2.28.0.241029 bash build/build_android_library.sh
+        PYTHON_EXECUTABLE=python ANDROID_ABIS="arm64-v8a" BUILD_AAR_DIR=aar-out EXECUTORCH_BUILD_QNN=ON QNN_SDK_ROOT=/tmp/qnn/2.28.0.241029 bash scripts/build_android_library.sh
         mkdir -p extension/benchmark/android/benchmark/app/libs
         cp aar-out/executorch.aar extension/benchmark/android/benchmark/app/libs
         pushd extension/benchmark/android/benchmark