Update on "Reuse types in _named_data_store and support tensor layouts"

Reuse `DataEntry` from data_serializer.py, in _named_data_store.py.

Motivation
- deserialize from flat tensor to named data store output
- support tensor layout in named data store

Differential Revision: [D83490345](https://our.internmc.facebook.com/intern/diff/D83490345/)

[ghstack-poisoned]

Author: lucylq

.ci/scripts/build-qnn-sdk.sh

@@ -18,7 +18,7 @@ build_qnn_backend() {
   export EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd)"
 
   parallelism=$(( $(nproc) - 1 ))
-  bash backends/qualcomm/scripts/build.sh --skip_aarch64 --job_number ${parallelism} --release
+  bash backends/qualcomm/scripts/build.sh --skip_linux_android --skip_linux_embedded --job_number ${parallelism} --release
 }
 
 set_up_aot() {

.github/workflows/_unittest.yml

@@ -32,7 +32,7 @@ jobs:
       id-token: write
       contents: read
     with:
-      runner: linux.2xlarge
+      runner: linux.2xlarge.memory
       docker-image: ${{ inputs.docker-image }}
       submodules: 'recursive'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}

.github/workflows/cuda.yml

@@ -128,7 +128,7 @@ jobs:
         echo "::endgroup::"
 
         echo "::group::Setup Huggingface"
-        pip install -U "huggingface_hub[cli]" accelerate
+        pip install -U "huggingface_hub[cli]<1.0" accelerate
         huggingface-cli login --token $SECRET_EXECUTORCH_HF_TOKEN
         OPTIMUM_ET_VERSION=$(cat .ci/docker/ci_commit_pins/optimum-executorch.txt)
         pip install git+https://github.com/huggingface/optimum-executorch.git@${OPTIMUM_ET_VERSION}
@@ -208,7 +208,7 @@ jobs:
         echo "::endgroup::"
 
         echo "::group::Setup Huggingface"
-        pip install -U "huggingface_hub[cli]" accelerate
+        pip install -U "huggingface_hub[cli]<1.0" accelerate
         huggingface-cli login --token $SECRET_EXECUTORCH_HF_TOKEN
         OPTIMUM_ET_VERSION=$(cat .ci/docker/ci_commit_pins/optimum-executorch.txt)
         pip install git+https://github.com/huggingface/optimum-executorch.git@${OPTIMUM_ET_VERSION}

.github/workflows/metal.yml

@@ -30,6 +30,8 @@ jobs:
 
   export-voxtral-metal-artifact:
     name: export-voxtral-metal-artifact
+      # Skip this job if the pull request is from a fork (HuggingFace secrets are not available)
+    if: github.event.pull_request.head.repo.full_name == github.repository || github.event_name != 'pull_request'
     uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
     secrets: inherit
     with:
@@ -44,7 +46,7 @@ jobs:
         set -eux
 
         echo "::group::Setup Huggingface"
-        ${CONDA_RUN} pip install -U "huggingface_hub[cli]" accelerate
+        ${CONDA_RUN} pip install -U "huggingface_hub[cli]<1.0" accelerate
         ${CONDA_RUN} huggingface-cli login --token $SECRET_EXECUTORCH_HF_TOKEN
         echo "::endgroup::"
 

.github/workflows/pull.yml

@@ -315,7 +315,7 @@ jobs:
         echo "::endgroup::"
 
         echo "::group::Setup Huggingface"
-        pip install -U "huggingface_hub[cli]" accelerate
+        pip install -U "huggingface_hub[cli]<1.0" accelerate
         huggingface-cli login --token $SECRET_EXECUTORCH_HF_TOKEN
         OPTIMUM_ET_VERSION=$(cat .ci/docker/ci_commit_pins/optimum-executorch.txt)
         pip install git+https://github.com/huggingface/optimum-executorch.git@${OPTIMUM_ET_VERSION}
@@ -636,7 +636,7 @@ jobs:
         echo "::group::Setup ExecuTorch"
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh --build-tool "cmake"
         echo "::endgroup::"
-                
+
         echo "::group::Setup requirements"
         # install phi-3-mini requirements
         bash examples/models/phi-3-mini/install_requirements.sh
@@ -909,6 +909,8 @@ jobs:
 
   test-samsung-models-linux:
     name: test-samsung-models-linux
+    # Skip this job if the pull request is from a fork (secrets are not available)
+    if: github.event.pull_request.head.repo.full_name == github.repository || github.event_name != 'pull_request'
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     permissions:
       id-token: write

.github/workflows/trunk.yml

@@ -626,7 +626,7 @@ jobs:
           BUILD_TORCHAO_EXPERIMENTAL=1 TORCHAO_BUILD_CPU_AARCH64=1 TORCHAO_BUILD_KLEIDIAI=1 TORCHAO_ENABLE_ARM_NEON_DOT=1 TORCHAO_PARALLEL_BACKEND=OPENMP pip install third-party/ao
         fi
 
-        pip install -U "huggingface_hub[cli]"
+        pip install -U "huggingface_hub[cli]<1.0"
 
         bash .ci/scripts/test_torchao_huggingface_checkpoints.sh ${{ matrix.model }} ${{ matrix.model != 'phi_4_mini' && '--test_with_runner' || '' }}  ${{ matrix.backend == 'torchao' && '--use_torchao_kernels' || '' }}
 
@@ -659,7 +659,7 @@ jobs:
         echo "::endgroup::"
 
         echo "::group::Set up Huggingface"
-        ${CONDA_RUN} pip install -U "huggingface_hub[cli]" accelerate
+        ${CONDA_RUN} pip install -U "huggingface_hub[cli]<1.0" accelerate
         ${CONDA_RUN} huggingface-cli login --token $SECRET_EXECUTORCH_HF_TOKEN
         OPTIMUM_ET_VERSION=$(cat .ci/docker/ci_commit_pins/optimum-executorch.txt)
         ${CONDA_RUN} pip install git+https://github.com/huggingface/optimum-executorch.git@${OPTIMUM_ET_VERSION}
@@ -834,7 +834,7 @@ jobs:
         echo "::endgroup::"
 
         echo "::group::Setup Huggingface"
-        pip install -U "huggingface_hub[cli]" accelerate
+        pip install -U "huggingface_hub[cli]<1.0" accelerate
         huggingface-cli login --token $SECRET_EXECUTORCH_HF_TOKEN
         OPTIMUM_ET_VERSION=$(cat .ci/docker/ci_commit_pins/optimum-executorch.txt)
         pip install git+https://github.com/huggingface/optimum-executorch.git@${OPTIMUM_ET_VERSION}
@@ -930,7 +930,7 @@ jobs:
         echo "::endgroup::"
 
         echo "::group::Set up Huggingface"
-        ${CONDA_RUN} pip install -U "huggingface_hub[cli]" accelerate
+        ${CONDA_RUN} pip install -U "huggingface_hub[cli]<1.0" accelerate
         ${CONDA_RUN} huggingface-cli login --token $SECRET_EXECUTORCH_HF_TOKEN
         OPTIMUM_ET_VERSION=$(cat .ci/docker/ci_commit_pins/optimum-executorch.txt)
         ${CONDA_RUN} pip install git+https://github.com/huggingface/optimum-executorch.git@${OPTIMUM_ET_VERSION}
@@ -1043,45 +1043,6 @@ jobs:
       build-tool: cmake
       docker-image: ci-image:executorch-ubuntu-22.04-clang12
 
-  test-mcu-models:
-    name: test-mcu-models
-    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
-    strategy:
-      matrix:
-        include:
-          - build-tool: cmake
-      fail-fast: false
-    permissions:
-      id-token: write
-      contents: read
-    with:
-      runner: linux.2xlarge
-      docker-image: ci-image:executorch-ubuntu-22.04-arm-sdk
-      submodules: 'recursive'
-      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-      timeout: 90
-      script: |
-        BUILD_TOOL=${{ matrix.build-tool }}
-
-        # The generic Linux job chooses to use base env, not the one setup by the image
-        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
-        conda activate "${CONDA_ENV}"
-
-        # Try to mirror these as closely as possible
-        source .ci/scripts/utils.sh
-        install_executorch "--use-pt-pinned-commit"
-
-        .ci/scripts/setup-arm-baremetal-tools.sh
-        source examples/arm/ethos-u-scratch/setup_path.sh
-
-        # Run selective Build
-        chmod +x examples/selective_build/test_selective_build.sh
-        examples/selective_build/test_selective_build.sh "${BUILD_TOOL}"
-
-        # Run MCU models
-        chmod +x examples/arm/run_mcu_models_fvp.sh
-        examples/arm/run_mcu_models_fvp.sh --target=cortex-m55
-
   test-models-windows:
     uses: pytorch/test-infra/.github/workflows/windows_job.yml@main
     strategy:

backends/aoti/common_shims.cpp

@@ -184,6 +184,10 @@ int32_t aoti_torch_dtype_int32() {
   return 3; // PyTorch's int32 dtype code
 }
 
+int32_t aoti_torch_dtype_bool() {
+  return 11; // PyTorch's bool dtype code
+}
+
 int32_t aoti_torch_dtype_int64() {
   return 4; // PyTorch's int64 dtype code
 }

backends/aoti/common_shims.h

@@ -63,6 +63,7 @@ int32_t aoti_torch_dtype_int8();
 int32_t aoti_torch_dtype_int16();
 int32_t aoti_torch_dtype_int32();
 int32_t aoti_torch_dtype_int64();
+int32_t aoti_torch_dtype_bool();
 
 // Dtype utility function needed by Metal backend
 size_t aoti_torch_dtype_element_size(int32_t dtype);

backends/aoti/tests/test_common_shims.cpp

@@ -322,3 +322,14 @@ TEST_F(CommonShimsTest, IndependentCaches) {
   // Sizes and strides pointers should be different (different caches)
   EXPECT_NE(sizes_ptr1, strides_ptr1);
 }
+
+// Test all dtype functions return correct PyTorch dtype codes
+TEST_F(CommonShimsTest, AllDtypesReturnCorrectValues) {
+  EXPECT_EQ(aoti_torch_dtype_float32(), 6); // PyTorch's float32 dtype code
+  EXPECT_EQ(aoti_torch_dtype_bfloat16(), 15); // PyTorch's bfloat16 dtype code
+  EXPECT_EQ(aoti_torch_dtype_int8(), 1); // PyTorch's int8 dtype code
+  EXPECT_EQ(aoti_torch_dtype_int16(), 2); // PyTorch's int16 dtype code
+  EXPECT_EQ(aoti_torch_dtype_int32(), 3); // PyTorch's int32 dtype code
+  EXPECT_EQ(aoti_torch_dtype_int64(), 4); // PyTorch's int64 dtype code
+  EXPECT_EQ(aoti_torch_dtype_bool(), 11); // PyTorch's bool dtype code
+}

backends/aoti/utils.h

@@ -45,6 +45,8 @@ inline executorch::aten::ScalarType dtype_to_scalar_type(int32_t dtype) {
       return executorch::aten::ScalarType::Long;
     case 6: // PyTorch's float32 dtype code
       return executorch::aten::ScalarType::Float;
+    case 11: // PyTorch's bool dtype code
+      return executorch::aten::ScalarType::Bool;
     case 15: // PyTorch's bfloat16 dtype code
       return executorch::aten::ScalarType::BFloat16;
     // Future support for additional dtypes can be added here