Restructing demos - aot

.ci/scripts/gather_test_models.py

@@ -10,7 +10,7 @@
 from typing import Any
 
 from examples.models import MODEL_NAME_TO_MODEL
-from examples.recipes.xnnpack_optimization import MODEL_NAME_TO_OPTIONS
+from examples.recipes.xnnpack import MODEL_NAME_TO_OPTIONS
 
 BUILD_TOOLS = [
     "buck2",
@@ -51,7 +51,8 @@ def export_models_for_ci() -> None:
     for name in MODEL_NAME_TO_MODEL.keys():
         quantization_configs = {
             False,
-            name in MODEL_NAME_TO_OPTIONS and MODEL_NAME_TO_OPTIONS[name].quantization,
+            name in MODEL_NAME_TO_OPTIONS
+            and MODEL_NAME_TO_OPTIONS[name].xnnpack_quantization,
         }
         delegation_configs = {
             False,

.ci/scripts/test.sh

@@ -53,11 +53,11 @@ build_cmake_executor_runner() {
 }
 
 test_model() {
-  "${PYTHON_EXECUTABLE}" -m examples.export.export_example --model_name="${MODEL_NAME}"
+  "${PYTHON_EXECUTABLE}" -m examples.export.portable --model_name="${MODEL_NAME}"
 
   # Run test model
   if [[ "${BUILD_TOOL}" == "buck2" ]]; then
-    buck2 run //examples/executor_runner:executor_runner -- --model_path "./${MODEL_NAME}.pte"
+    buck2 run //examples/runtime/portable:executor_runner -- --model_path "./${MODEL_NAME}.pte"
   elif [[ "${BUILD_TOOL}" == "cmake" ]]; then
     if [[ ! -f ${CMAKE_OUTPUT_DIR}/executor_runner ]]; then
       build_cmake_executor_runner
@@ -92,24 +92,24 @@ test_model_with_xnnpack() {
 
   # Quantization-only
   if [[ ${WITH_QUANTIZATION} == true ]] && [[ ${WITH_DELEGATION} == false ]]; then
-    bash examples/quantization/test_quantize.sh "${BUILD_TOOL}" "${MODEL_NAME}"
+    bash examples/quantization/quant_flow/test_quantize.sh "${BUILD_TOOL}" "${MODEL_NAME}"
     exit 0
   fi
 
   # Delegation
   if [[ ${WITH_QUANTIZATION} == true ]]; then
     SUFFIX="q8"
-    "${PYTHON_EXECUTABLE}" -m examples.backend.xnnpack_examples --model_name="${MODEL_NAME}" --delegate --quantize
+    "${PYTHON_EXECUTABLE}" -m examples.recipes.xnnpack.aot_compiler --model_name="${MODEL_NAME}" --delegate --quantize
   else
     SUFFIX="fp32"
-    "${PYTHON_EXECUTABLE}" -m examples.backend.xnnpack_examples --model_name="${MODEL_NAME}" --delegate
+    "${PYTHON_EXECUTABLE}" -m examples.recipes.xnnpack.aot_compiler --model_name="${MODEL_NAME}" --delegate
   fi
 
   OUTPUT_MODEL_PATH="${MODEL_NAME}_xnnpack_${SUFFIX}.pte"
 
   # Run test model
   if [[ "${BUILD_TOOL}" == "buck2" ]]; then
-    buck2 run //examples/backend:xnn_executor_runner -- --model_path "${OUTPUT_MODEL_PATH}"
+    buck2 run //examples/runtime/xnnpack:xnn_executor_runner -- --model_path "${OUTPUT_MODEL_PATH}"
   elif [[ "${BUILD_TOOL}" == "cmake" ]]; then
     if [[ ! -f ${CMAKE_OUTPUT_DIR}/backends/xnnpack/xnn_executor_runner ]]; then
       build_cmake_xnn_executor_runner
@@ -123,15 +123,15 @@ test_model_with_xnnpack() {
 
 test_demo_backend_delegation() {
   echo "Testing demo backend delegation on AddMul"
-  "${PYTHON_EXECUTABLE}" -m examples.export.export_and_delegate  --option "composite"
-  "${PYTHON_EXECUTABLE}" -m examples.export.export_and_delegate  --option "partition"
-  "${PYTHON_EXECUTABLE}" -m examples.export.export_and_delegate  --option "whole"
+  "${PYTHON_EXECUTABLE}" -m examples.recipes.export_and_delegate  --option "composite"
+  "${PYTHON_EXECUTABLE}" -m examples.recipes.export_and_delegate  --option "partition"
+  "${PYTHON_EXECUTABLE}" -m examples.recipes.export_and_delegate  --option "whole"
 
   # Run test model
   if [[ "${BUILD_TOOL}" == "buck2" ]]; then
-    buck2 run //examples/executor_runner:executor_runner -- --model_path "./composite_model.pte"
-    buck2 run //examples/executor_runner:executor_runner -- --model_path "./partition_lowered_model.pte"
-    buck2 run //examples/executor_runner:executor_runner -- --model_path "./whole.pte"
+    buck2 run //examples/runtime/portable:executor_runner -- --model_path "./composite_model.pte"
+    buck2 run //examples/runtime/portable:executor_runner -- --model_path "./partition_lowered_model.pte"
+    buck2 run //examples/runtime/portable:executor_runner -- --model_path "./whole.pte"
   elif [[ "${BUILD_TOOL}" == "cmake" ]]; then
     if [[ ! -f ${CMAKE_OUTPUT_DIR}/executor_runner ]]; then
       build_cmake_executor_runner

.ci/scripts/utils.sh

@@ -70,7 +70,7 @@ install_flatc_from_source() {
 
 build_executorch_runner_buck2() {
   # Build executorch runtime with retry as this step is flaky on macos CI
-  retry buck2 build //examples/executor_runner:executor_runner
+  retry buck2 build //examples/runtime/portable:executor_runner
 }
 
 build_executorch_runner_cmake() {

build/cmake_deps.toml

@@ -13,7 +13,7 @@ excludes = [
 
 [targets.executor_runner]
 buck_targets = [
-  "//examples/executor_runner:executor_runner",
+  "//examples/runtime/portable:executor_runner",
 ]
 filters = [
   ".cpp$",
@@ -93,7 +93,7 @@ deps = [
 
 [targets.xnn_executor_runner]
 buck_targets = [
-  "//examples/backend:xnn_executor_runner",
+  "//examples/runtime/xnnpack:xnn_executor_runner",
 ]
 filters = [
   ".cpp$",

docs/source/getting-started-setup.md

@@ -105,17 +105,17 @@ how to do it, we will generate an ExecuTorch program file from an `nn.Module`.
 You can generate an ExecuTorch program by using a sample script or by using
 the Python interpreter.
 
-We have created the `export_example.py` script that demonstrates a simple model
+We have created the `portable.py` script that demonstrates a simple model
 export to flatbuffer. This script is available
 in the [pytorch/executorch](https://github.com/pytorch/executorch/tree/main/examples/export)
 repository.
 
 To generate a sample program, complete the following steps:
 
-1. Run the `export_example.py` script:
+1. Run the `portable.py` script:
 
   ```bash
-  python3 -m examples.export.export_example --model_name="add"
+  python3 -m examples.export.portable --model_name="add"
   ```
 
   :::{dropdown} Output
@@ -193,7 +193,7 @@ Complete the following steps:
 
 3. Build a binary:
    ```bash
-   /tmp/buck2 build //examples/executor_runner:executor_runner --show-output
+   /tmp/buck2 build //examples/runtime/portable:executor_runner --show-output
    ```
 
    :::{dropdown} Output
@@ -227,7 +227,7 @@ the `buck run` command to run our program.
    * To run the `add.pte` program:
 
      ```bash
-     /tmp/buck2 run //examples/executor_runner:executor_runner -- --model_path add.pte
+     /tmp/buck2 run //examples/runtime/portable:executor_runner -- --model_path add.pte
      ```
 
      :::{dropdown} Sample Output

docs/website/docs/tutorials/00_setting_up_executorch.md

@@ -58,10 +58,10 @@ corresponsing version of the repo.
 Via python script:
 ```bash
 # Creates the file `add.pte`
-python3 -m examples.export.export_example --model_name="add"
+python3 -m examples.export.portable --model_name="add"
 
 # Creates the delegated program `composite_model.pte`, other options are "whole" and "partition"
-python3 -m examples.export.export_and_delegate --option "composite"
+python3 -m examples.recipes.export_and_delegate --option "composite"
 ```
 
 Or via python interpreter:
@@ -98,7 +98,7 @@ You may want to copy the `buck2` binary into your `$PATH` so you can run it as `
 `executor_runner` is an example wrapper around executorch runtime which includes all the operators and backends
 
 ```bash
-/tmp/buck2 build //examples/executor_runner:executor_runner --show-output
+/tmp/buck2 build //examples/runtime/portable:executor_runner --show-output
 ```
 
 The `--show-output` flag will print the path to the executable if you want to run it directly.
@@ -112,10 +112,10 @@ conda install -c conda-forge lld
 
 ```bash
 # add.pte is the program generated from export_example.py during AOT Setup Step 3
-/tmp/buck2 run //examples/executor_runner:executor_runner -- --model_path add.pte
+/tmp/buck2 run //examples/runtime/portable:executor_runner -- --model_path add.pte
 
 # To run a delegated model
-/tmp/buck2 run //examples/executor_runner:executor_runner -- --model_path composite_model.pte
+/tmp/buck2 run //examples/runtime/portable:executor_runner -- --model_path composite_model.pte
 ```
 
 or execute the binary directly from the `--show-output` path shown when building.

docs/website/docs/tutorials/profiling.md

@@ -56,7 +56,7 @@ Here is an example of what a ExecuTorch run + profile + post-procesing workflow
 This runs the sample program with profiling enabled
 ```bash
 cd executorch
-buck2 run -c executorch.prof_enabled=true examples/executor_runner:executor_runner  -- --model_path add.pte
+buck2 run -c executorch.prof_enabled=true examples/runtime/portable:executor_runner  -- --model_path add.pte
 ```
 Run the post-processing CLI tool that calls into the same API's listed above and prints out the profiling results in a tabulated format in the terminal.
 

examples/README.md

@@ -1,6 +1,6 @@
 # Examples
 
-This dir contains scripts and other helper utilities to illustrate an end-to-end workflow to run a torch.nn.module on the ExecuTorch runtime.
+This dir contains scripts and other helper utilities to illustrate an end-to-end workflow to run a torch.nn.module on the Executorch runtime.
 It also includes a list of modules, from a simple `Add` to a full model like `MobileNetv2` and `MobileNetv3`, with more to come.
 
 
@@ -10,12 +10,12 @@ examples
 |── backend                           # Contains examples for exporting delegate models and running them using custom executor runners
 ├── custom_ops                        # Contains examples to register custom operators into PyTorch as well as register its kernels into ExecuTorch runtime
 ├── example_quantizer_and_delegate    # Contains examples to to fully lowered a MobileNetV2 model to the example backend with an example quantizer
-├── executor_runner                   # This is an example C++ wrapper around the ET runtime
-├── export                            # Python helper scripts to illustrate export workflow
+├── export                            # Contains scripts to illustrate export workflow in portable mode
 ├── ios_demo_apps                     # Contains iOS demo apps
-├── models                            # Contains a set of simple to PyTorch models
+├── models                            # Contains a set of out-of-box PyTorch models
 ├── quantization                      # Contains examples of quantization workflow
-├── arm                               # Contains examples of the Arm TOSA and Ethos-U NPU flows
+├── recipes                           # Contains recipes for a set of demos
+├── runtime                           # Contains examples of C++ wrapper around the ET runtime
 └── README.md                         # This file
 ```
 
@@ -31,18 +31,18 @@ and executing previously exported binary file(s).
 1. Following the setup guide in [Setting up ExecuTorch from GitHub](/docs/website/docs/tutorials/00_setting_up_executorch.md)
 you should be able to get the basic development environment for ExecuTorch working.
 
-2. Using the script `export/export_example.py` generate a model binary file by selecting a
+2. Using the example script `portable/aot_compiler.py` generate a model binary file by selecting a
 model name from the list of available models in the `models` dir.
 
 
 ```bash
 cd executorch # To the top level dir
 
 # To get a list of example models
-python3 -m examples.export.export_example -h
+python3 -m examples.export.portable -h
 
 # To generate a specific pte model
-python3 -m examples.export.export_example --model_name="mv2" # for MobileNetv2
+python3 -m examples.export.portable --model_name="mv2" # for MobileNetv2
 
 # This should generate ./mv2.pte file, if successful.
 ```
@@ -52,7 +52,7 @@ Use `-h` (or `--help`) to see all the supported models.
 3. Once we have the model binary (pte) file, then let's run it with ExecuTorch runtime using the `executor_runner`.
 
 ```bash
-buck2 run examples/executor_runner:executor_runner -- --model_path mv2.pte
+buck2 run examples/runtime/portable:executor_runner -- --model_path mv2.pte
 ```
 
 ## Quantization
@@ -87,7 +87,7 @@ buck2 run executorch/examples/quantization:example -- --help
 Quantized model can be run via executor_runner, similar to floating point model, via, as shown above:
 
 ```bash
-buck2 run examples/executor_runner:executor_runner -- --model_path mv2.pte
+buck2 run examples/runtime/portable:executor_runner -- --model_path mv2.pte
 ```
 
 Note that, running quantized model, requires various quantized/dequantize operators, available in [quantized kernel lib](/kernels/quantized).

examples/custom_ops/custom_ops_1.py

@@ -6,7 +6,7 @@
 
 """Example of showcasing registering custom operator through torch library API."""
 import torch
-from examples.export.export_example import export_to_exec_prog, save_pte_program
+from examples.export.utils import export_to_exec_prog, save_pte_program
 
 from executorch.exir import EdgeCompileConfig
 from torch.library import impl, Library

examples/custom_ops/custom_ops_2.py

@@ -11,7 +11,7 @@
 import argparse
 
 import torch
-from examples.export.export_example import export_to_exec_prog, save_pte_program
+from examples.export.utils import export_to_exec_prog, save_pte_program
 from executorch.exir import EdgeCompileConfig
 
 

examples/custom_ops/test_custom_ops.sh

@@ -21,7 +21,7 @@ test_buck2_custom_op_1() {
   # should save file custom_ops_1.pte
 
   echo 'Running executor_runner'
-  buck2 run //examples/executor_runner:executor_runner \
+  buck2 run //examples/runtime/portable:executor_runner \
       --config=executorch.register_custom_op=1 -- --model_path="./${model_name}.pte"
   # should give correct result
 
@@ -58,7 +58,7 @@ test_buck2_custom_op_2() {
   ${PYTHON_EXECUTABLE} -m "examples.custom_ops.${model_name}" --so_library="$SO_LIB"
   # should save file custom_ops_2.pte
 
-  buck2 run //examples/executor_runner:executor_runner \
+  buck2 run //examples/runtime/portable:executor_runner \
       --config=executorch.register_custom_op=2 -- --model_path="./${model_name}.pte"
   # should give correct result
   echo "Removing ${model_name}.pte"

examples/export/export_example.py → examples/export/portable.py

@@ -11,6 +11,7 @@
 
 from ..models import MODEL_NAME_TO_MODEL
 from ..models.model_factory import EagerModelFactory
+
 from .utils import export_to_exec_prog, save_pte_program
 
 

examples/ios_demo_apps/README.md

@@ -20,7 +20,7 @@ and UI looks like
 
 ![](./executorch_mobilenet_ui.png)
 
-Step 1. Export a MobileNetV2 model following example follwing executorch/examples/export/export_example.py. Instead of export mv2 directly, add a softmax at the end
+Step 1. Export a MobileNetV2 model following example follwing executorch/examples/export/portable.py. Instead of export mv2 directly, add a softmax at the end
 ```python
 class MobileNetV2Wrapper(torch.nn.Module):
     def __init__(self):

examples/quantization/quant_flow/TARGETS

@@ -0,0 +1,26 @@
+load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
+
+runtime.python_binary(
+    name = "example",
+    main_src = "example.py",
+    preload_deps = ["//executorch/kernels/quantized:aot_lib"],
+    deps = [
+        ":quant_utils",
+        "//caffe2:torch",
+        "//executorch/examples/export:lib",
+        "//executorch/examples/recipes/xnnpack:models",
+    ],
+)
+
+runtime.python_library(
+    name = "quant_utils",
+    srcs = [
+        "utils.py",
+    ],
+    visibility = [
+        "//executorch/examples/...",
+    ],
+    deps = [
+        "//caffe2:torch",
+    ],
+)