Merge branch 'main' into jz/remove-ckpt-ci

Author: jackzhxng

.ci/scripts/test_model.sh

@@ -224,19 +224,22 @@ test_model_with_coreml() {
 
   "${PYTHON_EXECUTABLE}" -m examples.apple.coreml.scripts.export --model_name="${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"
+    echo "OK exported model: $EXPORTED_MODEL"
   else
-    echo "No .pte file found"
+    echo "[error] failed to export model: no .pte file found"
     exit 1
   fi
 
   # Run the model
   if [ "${should_test}" = true ]; then
+    echo "Installing requirements needed to build coreml_executor_runner..."
+    backends/apple/coreml/scripts/install_requirements.sh
+
     echo "Testing exported model with coreml_executor_runner..."
     local out_dir=$(mktemp -d)
     COREML_EXECUTOR_RUNNER_OUT_DIR="${out_dir}" examples/apple/coreml/scripts/build_executor_runner.sh

.ci/scripts/wheel/test_macos.py

@@ -15,11 +15,9 @@
                 model=Model.Mv3,
                 backend=Backend.XnnpackQuantizationDelegation,
             ),
-            # Enable this once CoreML is suppported out-of-the-box
-            # https://github.com/pytorch/executorch/issues/9019
-            # test_base.ModelTest(
-            #     model=Model.Mv3,
-            #     backend=Backend.CoreMlTest,
-            # )
+            test_base.ModelTest(
+                model=Model.Mv3,
+                backend=Backend.CoreMlTest,
+            ),
         ]
     )

.github/workflows/build-wheels-linux.yml

@@ -6,6 +6,9 @@ on:
     paths:
       - .ci/**/*
       - .github/workflows/build-wheels-linux.yml
+      - examples/**/*
+      - pyproject.toml
+      - setup.py
   push:
     branches:
       - nightly

.github/workflows/build-wheels-macos.yml

@@ -6,6 +6,9 @@ on:
     paths:
       - .ci/**/*
       - .github/workflows/build-wheels-macos.yml
+      - examples/**/*
+      - pyproject.toml
+      - setup.py
   push:
     branches:
       - nightly
@@ -57,6 +60,8 @@ jobs:
       pre-script: ${{ matrix.pre-script }}
       post-script: ${{ matrix.post-script }}
       package-name: ${{ matrix.package-name }}
-      runner-type: macos-m1-stable
+      # Meta's macOS runners do not have Xcode, so use GitHub's runners.
+      runner-type: macos-latest-xlarge
+      setup-miniconda: true
       smoke-test-script: ${{ matrix.smoke-test-script }}
       trigger-event: ${{ github.event_name }}

.github/workflows/trunk.yml

@@ -65,22 +65,29 @@ jobs:
       matrix:
         model: [linear, add, add_mul, ic3, ic4, mv2, mv3, resnet18, resnet50, vit, w2l, mobilebert, emformer_join, emformer_transcribe]
         backend: [portable, xnnpack-quantization-delegation]
+        runner: [linux.arm64.2xlarge]
         include:
           - model: lstm
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: mul
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: softmax
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: phi_4_mini
             backend: portable
+            runner: linux.arm64.m7g.4xlarge
           - model: qwen2_5
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: llama3_2_vision_encoder
             backend: portable
+            runner: linux.arm64.2xlarge
       fail-fast: false
     with:
-      runner: linux.arm64.2xlarge
+      runner: ${{ matrix.runner }}
       docker-image: executorch-ubuntu-22.04-gcc11-aarch64
       submodules: 'true'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
@@ -536,9 +543,8 @@ jobs:
         git clone https://github.com/huggingface/optimum-executorch
         cd optimum-executorch
         # There is no release yet, for CI stability, always test from the same commit on main
-        git checkout 6a7e83f3eee2976fa809335bfb78a45b1ea1cb25
-        pip install .
-        pip install accelerate sentencepiece
+        git checkout 577a2b19670e4c643a5c6ecb09bf47b9a699e7c6
+        pip install .[tests]
         pip list
         echo "::endgroup::"
 

backends/arm/test/conftest.py

@@ -44,7 +44,10 @@ def pytest_configure(config):
             )
         # Only enable if we also have the TOSA reference model available.
         pytest._test_options["corstone_fvp"] = True  # type: ignore[attr-defined]
-    pytest._test_options["llama_inputs"] = config.option.llama_inputs  # type: ignore[attr-defined]
+
+    if getattr(config.option, "llama_inputs", False) and config.option.llama_inputs:
+        pytest._test_options["llama_inputs"] = config.option.llama_inputs  # type: ignore[attr-defined]
+
     pytest._test_options["fast_fvp"] = False  # type: ignore[attr-defined]
     if getattr(config.option, "fast_fvp", False):
         pytest._test_options["fast_fvp"] = config.option.fast_fvp  # type: ignore[attr-defined]

backends/arm/test/misc/test_debug_feats.py

@@ -197,10 +197,10 @@ def test_collate_tosa_BI_tests(self):
             "test_collate_tosa_tests/tosa-bi/TestCollateTosaTests/test_collate_tosa_BI_tests"
         )
         assert os.path.exists(
-            "test_collate_tosa_tests/tosa-bi/TestCollateTosaTests/test_collate_tosa_BI_tests/output_tag6.tosa"
+            "test_collate_tosa_tests/tosa-bi/TestCollateTosaTests/test_collate_tosa_BI_tests/output_tag6_TOSA-0.80+BI.tosa"
         )
         assert os.path.exists(
-            "test_collate_tosa_tests/tosa-bi/TestCollateTosaTests/test_collate_tosa_BI_tests/desc_tag6.json"
+            "test_collate_tosa_tests/tosa-bi/TestCollateTosaTests/test_collate_tosa_BI_tests/desc_tag6_TOSA-0.80+BI.json"
         )
 
         os.environ.pop("TOSA_TESTCASES_BASE_PATH")

backends/arm/tosa_backend.py

@@ -125,7 +125,7 @@ def preprocess(  # noqa: C901
             dbg_tosa_dump(
                 tosa_graph,
                 artifact_path,
-                suffix="{}".format(f"_{tag}" if tag else ""),
+                suffix="{}".format(f"_{tag}" if tag else "") + (f"_{tosa_spec}"),
             )
 
         # Serialize and return the TOSA flatbuffer.

backends/vulkan/runtime/api/Context.cpp

@@ -272,7 +272,7 @@ Context* context() {
 
 VkPipeline Context::get_shader_pipeline(
     const vkapi::ShaderInfo& shader,
-    const vkapi::SpecVarList& spec_constants) {
+    const vkapi::SpecVarList& additional_constants) {
   const uint32_t push_constants_size = 128u;
 
   VkDescriptorSetLayout shader_layout =
@@ -281,12 +281,15 @@ VkPipeline Context::get_shader_pipeline(
       pipeline_layout_cache().retrieve(shader_layout, push_constants_size);
 
   const utils::WorkgroupSize local_workgroup_size(4u, 4u, 1u);
+  vkapi::SpecVarList spec_constants = {
+      SV(local_workgroup_size[0u]),
+      SV(local_workgroup_size[1u]),
+      SV(local_workgroup_size[2u])};
+
+  spec_constants.append(additional_constants);
 
   VkPipeline pipeline = pipeline_cache().retrieve(
-      {pipeline_layout,
-       shader_cache().retrieve(shader),
-       spec_constants,
-       local_workgroup_size});
+      {pipeline_layout, shader_cache().retrieve(shader), spec_constants});
 
   return pipeline;
 }

backends/vulkan/runtime/graph/ops/glsl/clone.glsl

@@ -8,16 +8,15 @@
 
 #version 450 core
 
+#include "indexing_utils.h"
+
 #define PRECISION ${PRECISION}
 
 layout(std430) buffer;
 
-layout(set = 0, binding = 0, ${IMAGE_FORMAT[DTYPE]}) uniform PRECISION restrict writeonly ${IMAGE_T[NDIM][DTYPE]} image_out;
-layout(set = 0, binding = 1) uniform PRECISION sampler3D image_in;
-
-layout(set = 0, binding = 2) uniform PRECISION restrict OutLimits {
-  ivec3 out_limits;
-};
+${layout_declare_tensor(B, "w", "t_out", DTYPE, STORAGE)}
+${layout_declare_tensor(B, "r", "t_in", DTYPE, STORAGE)}
+${layout_declare_ubo(B, "ivec3", "out_limits")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
@@ -26,5 +25,5 @@ void main() {
   if (any(greaterThanEqual(pos, out_limits))) {
     return;
   }
-  imageStore(image_out, pos, texelFetch(image_in, pos, 0));
+  imageStore(t_out, pos, load_texel(t_in, pos));
 }