### Installing additional Python packages

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/README.md

Command to install necessary Python packages for the examples.

```bash
pip install scikit-learn pandas graphviz
```

--------------------------------

### Install Transformers

Source: https://github.com/pytorch/executorch/blob/main/examples/models/phi-3-mini/README.md

Installs the required version of the transformers library.

```bash
pip uninstall -y transformers ; pip install transformers==4.56.1
```

--------------------------------

### Install optimum-executorch

Source: https://github.com/pytorch/executorch/blob/main/examples/models/phi-3-mini/README.md

Installs the optimum-executorch package from a specific commit.

```bash
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}
```

--------------------------------

### Getting Help for Quantized Example Models

Source: https://github.com/pytorch/executorch/blob/main/examples/xnnpack/README.md

Command to find more valid quantized example models.

```bash
python3 -m examples.xnnpack.quantization.example --help
```

--------------------------------

### Install Optimum ExecuTorch from source

Source: https://github.com/pytorch/executorch/blob/main/examples/models/gemma3/README.md

Installs the Optimum ExecuTorch library from its source repository.

```bash
git clone https://github.com/huggingface/optimum-executorch.git
cd optimum-executorch
python install_dev.py
```

--------------------------------

### CMakeLists.txt Configuration for ExecuTorch

Source: https://github.com/pytorch/executorch/blob/main/docs/source/getting-started.md

Example CMakeLists.txt configuration for integrating ExecuTorch into a C++ project.

```cmake
# Set CMAKE_CXX_STANDARD to 17 or above.
set(CMAKE_CXX_STANDARD 17)

# CMakeLists.txt
set(EXECUTORCH_BUILD_PRESET_FILE ${CMAKE_SOURCE_DIR}/executorch/tools/cmake/preset/llm.cmake)
# Set other ExecuTorch options here.

add_subdirectory("executorch")
...
target_link_libraries(
  my_target
  PRIVATE executorch
          executorch::backends
          executorch::extensions
          executorch::kernels)
```

--------------------------------

### Setup Arm toolchain and FVP

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/arm-cortex-m/arm-cortex-m-overview.md

One-time setup command to install the Arm toolchain and FVP.

```bash
# One-time: install the Arm toolchain + FVP.
examples/arm/setup.sh --i-agree-to-the-contained-eula
source examples/arm/arm-scratch/setup_path.sh
```

--------------------------------

### Include Directories Setup

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/CMakeLists.txt

Sets common include directories for the project.

```cmake
# Let files say "include <executorch/path/to/header.h>".
set(_common_include_directories
    ${EXECUTORCH_ROOT}/..
    ${EXECUTORCH_ROOT}/extension/llm/tokenizers/third-party/json/single_include
)
```

--------------------------------

### Running Ethos-U minimal example in a notebook

Source: https://github.com/pytorch/executorch/blob/main/examples/arm/README.md

Instructions to install Jupyter and run the Ethos-U minimal example notebook.

```bash
pip install jupyter
jupyter notebook ethos_u_minimal_example.ipynb
```

--------------------------------

### Install Optimum ExecuTorch via pip

Source: https://github.com/pytorch/executorch/blob/main/examples/models/gemma3/README.md

Installs the Optimum ExecuTorch library using pip.

```bash
pip install optimum-executorch
```

--------------------------------

### Environment Setup

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/template/tutorials/backend-basic-tutorial.md

Commands to create a Conda environment, activate it, and install the ExecuTorch Python package.

```bash
conda create -y --name executorch python=3.12
conda activate executorch
conda install executorch
```

--------------------------------

### Executorch and Dependencies Setup

Source: https://github.com/pytorch/executorch/blob/main/examples/models/llama/CMakeLists.txt

Configures Python executable, Executorch and PyTorch roots, and finds necessary packages like executorch and gflags.

```cmake
if(NOT PYTHON_EXECUTABLE)
  set(PYTHON_EXECUTABLE python3)
endif()

set(EXECUTORCH_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/../../..)
set(TORCH_ROOT ${EXECUTORCH_ROOT}/third-party/pytorch)

include(${EXECUTORCH_ROOT}/tools/cmake/Utils.cmake)

if(NOT PYTHON_EXECUTABLE)
  resolve_python_executable()
endif()

if(NOT CMAKE_CXX_STANDARD)
  set(CMAKE_CXX_STANDARD 17)
  # Can't set to 11 due to executor_runner.cpp make_unique
endif()

if(CMAKE_TOOLCHAIN_FILE MATCHES ".*(iOS|ios\.toolchain)\.cmake$")
  set(CMAKE_TOOLCHAIN_IOS ON)
else()
  set(CMAKE_TOOLCHAIN_IOS OFF)
endif()

set(_common_compile_options -Wno-deprecated-declarations -fPIC)

# Let files say "include <executorch/path/to/header.h>".
set(_common_include_directories ${EXECUTORCH_ROOT}/..)

# For some reason android build is not able to find where gflags is and hence
# cannot find corresponding .cmake file
set(gflags_DIR ${CMAKE_CURRENT_BINARY_DIR}/../../../third-party/gflags)
find_package(gflags REQUIRED)

#
# llama_main: test binary to run llama, with tokenizer and sampler integrated
#

# find `executorch` libraries. CMAKE_PREFIX_PATH would work for host
# compilation, but CMAKE_FIND_ROOT_PATH appears to be necessary for
# cross-compiling (e.g., to Android) to work as well.
list(APPEND CMAKE_FIND_ROOT_PATH ${CMAKE_CURRENT_BINARY_DIR}/../../..)
find_package(executorch CONFIG REQUIRED FIND_ROOT_PATH_BOTH)
executorch_target_link_options_shared_lib(executorch)
```

--------------------------------

### Project Setup and Variable Definitions

Source: https://github.com/pytorch/executorch/blob/main/examples/nxp/executor_runner/CMakeLists.txt

Initializes the CMake version, project name, and defines key variables like EIQ_NEUTRON_TARGET and EXECUTORCH_PATH.

```cmake
cmake_minimum_required(VERSION 3.16)
project(nxp_executor_runner C CXX)

set(EIQ_NEUTRON_TARGET
    "imxrt700"
    CACHE STRING "Neutron Target to build the executor_runner"
)
set(EXECUTORCH_PATH ${CMAKE_CURRENT_LIST_DIR}/../../../)
```

--------------------------------

### Quick Start Setup Script

Source: https://github.com/pytorch/executorch/blob/main/examples/riscv/README.md

Shell script to set up the necessary packages for RISC-V cross-compilation on Ubuntu/Debian.

```bash
examples/riscv/setup.sh        # apt: gcc-riscv64-linux-gnu, qemu-user-static
```

--------------------------------

### QWEN2.5 0.5B Default Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

Default example using hybrid mode for QWEN2.5 0.5B model.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 --decoder_model qwen2_5-0_5b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Install eIQ Neutron dependencies (again)

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/nxp/tutorials/nxp-basic-tutorial.md

Ensures the eIQ Neutron dependencies are installed.

```bash
$ ./examples/nxp/setup.sh
```

--------------------------------

### QWEN3 0.6B Default Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

Default example using hybrid mode for QWEN3 0.6B model.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 --decoder_model qwen3-0_6b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Install ExecuTorch Python package (prebuilt)

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/nxp/tutorials/nxp-basic-tutorial.md

Installs the ExecuTorch Python package using Conda.

```bash
conda install executorch
```

--------------------------------

### QWEN3 1.7B Default Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

Default example using hybrid mode for QWEN3 1.7B model.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --decoder_model qwen3-1_7b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Check Model Converter Installation

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/arm-vgf/tutorials/vgf-getting-started.md

A simple command to verify that the 'model-converter' executable is installed and accessible in the expected location within the 'examples/arm' directory.

```bash
which model-converter
```

--------------------------------

### QWEN2.5 1.5B Default Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

Default example using hybrid mode for QWEN2.5 1.5B model.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --decoder_model qwen2_5-1_5b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Install eIQ Neutron dependencies

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/nxp/tutorials/nxp-basic-tutorial.md

This command installs the Neutron Converter, eIQ Neutron SDK, and eIQ NSYS, which are necessary for working with the eIQ Neutron NPU.

```bash
examples/nxp/setup.sh
```

--------------------------------

### Example setup for CUDA-Windows export on Ubuntu

Source: https://github.com/pytorch/executorch/blob/main/examples/models/voxtral/README.md

Sets up the necessary cross-compiler and downloads/extracts the Windows CUDA installer package for cross-compilation.

```bash
# 1) Install cross-compiler + extraction tools
sudo apt-get update
sudo apt-get install -y --no-install-recommends \
  g++-mingw-w64-x86-64-posix mingw-w64-tools p7zip-full wget

# 2) Verify cross-compiler
x86_64-w64-mingw32-g++ --version

# 3) Download and extract Windows CUDA installer package
CUDA_VERSION=12.8.1
CUDA_DRIVER_VERSION=572.61
CUDA_INSTALLER="cuda_${CUDA_VERSION}_${CUDA_DRIVER_VERSION}_windows.exe"
CUDA_URL="https://developer.download.nvidia.com/compute/cuda/${CUDA_VERSION}/local_installers/${CUDA_INSTALLER}"

mkdir -p /opt/cuda-windows
cd /opt/cuda-windows
wget -q "${CUDA_URL}" -O "${CUDA_INSTALLER}"
7z x "${CUDA_INSTALLER}" -oextracted -y

# 4) Point WINDOWS_CUDA_HOME to extracted Windows CUDA payload
export WINDOWS_CUDA_HOME=/opt/cuda-windows/extracted/cuda_cudart/cudart
```

--------------------------------

### Static Backend Registration Example

Source: https://github.com/pytorch/executorch/blob/main/docs/source/compiler-delegate-and-partitioner.md

Example demonstrating static registration of a backend at library initialization or load time.

```cpp
namespace {
auto cls = BackendWithCompiler();
Backend backend{"BackendWithCompilerDemo", &cls};
static auto success_with_compiler = register_backend(backend);
} // namespace
```

--------------------------------

### Install Requirements

Source: https://github.com/pytorch/executorch/blob/main/examples/models/phi-3-mini-lora/README.md

Installs Python requirements for the Phi-3-mini-LoRA example.

```shell
./examples/models/phi-3-mini-lora/install_requirements.sh
```

--------------------------------

### Install VGF Dependencies

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/arm-vgf/tutorials/vgf-getting-started.md

This command installs necessary SDK dependencies for generating VGF files, including TOSA Serialization Library, ML SDK Model Converter, Vulkan API, and ML Emulation Layer for Vulkan.

```bash
./examples/arm/setup.sh --i-agree-to-the-contained-eula --disable-ethos-u-deps --enable-mlsdk-deps
```

--------------------------------

### Install ExecuTorch using uv

Source: https://github.com/pytorch/executorch/blob/main/extension/training/examples/CIFAR/README.md

Commands to clone the ExecuTorch repository, set up a Python environment with uv, and install ExecuTorch.

```bash
git clone https://github.com/pytorch/executorch.git --recurse-submodules
cd executorch
uv venv --seed --prompt et --python 3.10
source .venv/bin/activate
git fetch origin
git submodule sync --recursive
git submodule update --init --recursive
./install_executorch.sh
```

--------------------------------

### NeutronDriver API example

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/nxp/tutorials/nxp-basic-tutorial.md

Illustrates how to set paths for the NSYS simulator, Neutron Firmware, and NSYS configuration using the NeutronDriver API.

```c++
storeNsysConfigPath(FLAGS_nsys_config.c_str());
storeFirmwarePath(FLAGS_firmware.c_str());
storeNsysPath(FLAGS_nsys.c_str());
```

--------------------------------

### Run Gemma3 1B Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

This command executes the default example for Gemma3 1B using hybrid mode.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 --decoder_model gemma3-1b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Project and Options Configuration

Source: https://github.com/pytorch/executorch/blob/main/examples/models/llama/CMakeLists.txt

Sets up the minimum CMake version, project name, and options for building optimized kernels and libraries like pthreadpool and cpuinfo.

```cmake
cmake_minimum_required(VERSION 3.24) # 3.24 is required for WHOLE_ARCHIVE
project(llama_runner)

# Duplicating options as root CMakeLists.txt
option(EXECUTORCH_BUILD_KERNELS_OPTIMIZED "Build the optimized kernels" OFF)

include(CMakeDependentOption)
#
# pthreadpool: build pthreadpool library. Disable on unsupported platforms
#
cmake_dependent_option(
  EXECUTORCH_BUILD_PTHREADPOOL "Build pthreadpool library." ON
  "NOT EXECUTORCH_BUILD_ARM_BAREMETAL" OFF
)
#
# cpuinfo: build cpuinfo library. Disable on unsupported platforms
#
cmake_dependent_option(
  EXECUTORCH_BUILD_CPUINFO "Build cpuinfo library." ON
  "NOT EXECUTORCH_BUILD_ARM_BAREMETAL" OFF
)

option(EXECUTORCH_BUILD_KERNELS_TORCHAO_MPS "Build the torchao MPS kernels" OFF)
```

--------------------------------

### Requirements

Source: https://github.com/pytorch/executorch/blob/main/backends/mlx/examples/llm/README.md

Installs necessary Python packages for the example.

```bash
pip install transformers optimum-executorch
```

--------------------------------

### Run LLAMA3.2 1B Instruct Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

This command executes the default example for LLAMA3.2 1B Instruct using hybrid mode.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --checkpoint consolidated.00.pth --params params.json --tokenizer_model tokenizer.model --decoder_model llama3_2-1b_instruct --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Install Executorch Dependencies

Source: https://github.com/pytorch/executorch/blob/main/examples/demo-apps/react-native/rnllama/README.md

Installs necessary dependencies for Executorch.

```bash
./install_executorch.sh && ./examples/models/llama/install_requirements.sh
```

--------------------------------

### Running example scripts with shared buffer

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/README.md

Command to run example scripts with shared buffer enabled for inference.

```bash
python mobilenet_v2.py -s <device_serial> -m "SM8550" -b path/to/build-android/ -d /path/to/imagenet-mini/val --shared_buffer
```

--------------------------------

### AOT Compilation Example

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/arm-vgf/tutorials/vgf-getting-started.md

This Python code demonstrates how to quantize a simple PyTorch model (AddSigmoid) and export it through the AOT flow using the VGF backend. It includes model definition, input preparation, export, and quantization configuration using ExecuTorch and TorchAO libraries.

```python
import torch

class AddSigmoid(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.sigmoid = torch.nn.Sigmoid()

    def forward(self, x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
        return self.sigmoid(x + y)

example_inputs = (torch.ones(1,1,1,1),torch.ones(1,1,1,1))

model = AddSigmoid()
model = model.eval()
exported_program = torch.export.export(model, example_inputs)
graph_module = exported_program.module(check_guards=False)


from executorch.backends.arm.quantizer import (
    VgfQuantizer,
    get_symmetric_quantization_config,
)
from executorch.backends.arm.vgf import VgfCompileSpec
from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e

# Create a compilation spec describing the target for configuring the quantizer
compile_spec = VgfCompileSpec()

# Create and configure quantizer to use a symmetric quantization config globally on all nodes
quantizer = VgfQuantizer(compile_spec)
operator_config = get_symmetric_quantization_config(is_per_channel=False)

# Set global (default) quantization config for the layers in the models.
# Can also be set to `None` to let layers run in FP as default.
quantizer.set_global(operator_config)

# Skip quantizing all sigmoid ops (only one for this model); let it run in FP.
# This step is optional; selecting which layers to include/exclude for
# quantization is part of optimizing the model's performance.
quantizer.set_module_type(torch.nn.Sigmoid, None)

```

--------------------------------

### Main CMake Configuration

Source: https://github.com/pytorch/executorch/blob/main/examples/models/voxtral_realtime/CMakeLists.txt

Sets up the minimum CMake version, project name, C++ standard, and root directory for ExecuTorch.

```cmake
cmake_minimum_required(VERSION 3.24)
project(voxtral_realtime)

set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED ON)

set(EXECUTORCH_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/../../..)

include(${EXECUTORCH_ROOT}/tools/cmake/Utils.cmake)
```

--------------------------------

### Install RN Dependencies

Source: https://github.com/pytorch/executorch/blob/main/examples/demo-apps/react-native/rnllama/README.md

Installs Node.js and iOS dependencies for the React Native app.

```bash
yarn && cd ios && pod install && cd ..
```

--------------------------------

### Executorch Installation and Build Setup

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends-cadence.md

Cleans the Executorch installation and creates a directory for CMake output.

```bash
cd executorch
./install_executorch.sh --clean
mkdir cmake-out
```

--------------------------------

### Python Installation Options

Source: https://github.com/pytorch/executorch/blob/main/extension/llm/runner/README.md

Code examples for installing ExecuTorch with Python bindings, including options using the install script, CMake, and pip.

```bash
# Option 1: Use the install script (includes pybindings by default)
bash install_executorch.sh

# Option 2: Build with CMake directly
cmake -B cmake-out \
  -DEXECUTORCH_BUILD_PYBIND=ON \
  -DCMAKE_INSTALL_PREFIX=cmake-out \
  cmake/
cmake --build cmake-out -j$(nproc) --target install

# Option 3: pip install from source (includes pybindings)
pip install -e . --no-build-isolation
```

--------------------------------

### Run Gemma2 2B Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

This command executes the default example for Gemma2 2B using hybrid mode.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 --decoder_model gemma2-2b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Running on Corstone FVP Platforms

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/arm-ethos-u/tutorials/ethos-u-getting-started.md

This command demonstrates how to use a utility script to run the compiled executable (.elf-file) on a simulated Arm hardware environment (FVP).

```bash
backends/arm/scripts/run_fvp.sh --elf=$(find ethos_u_minimal_example -name arm_executor_runner) --target=ethos-u55-128
```

--------------------------------

### Build Core ML delegate frameworks

Source: https://github.com/pytorch/executorch/blob/main/backends/apple/coreml/setup.md

Builds the executorch.xcframework and coreml_backend.xcframework using a script.

```bash
cd executorch
./scripts/build_apple_frameworks.sh --coreml

```

--------------------------------

### Run LLAMA2 Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

This command executes the default example for LLAMA2 using hybrid mode.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --checkpoint stories110M.pt --params params.json --tokenizer_model tokenizer.model --tokenizer_bin tokenizer.bin --decoder_model stories110m --model_mode hybrid --prefill_ar_len 32 --max_seq_len 128 --prompt "Once upon a time"
```

--------------------------------

### Setup and install ExecuTorch

Source: https://github.com/pytorch/executorch/blob/main/zephyr/README.md

Navigates to the ExecuTorch directory, syncs and updates its submodules, and runs the installation script.

```bash
cd modules/lib/executorch/
git submodule sync
git submodule update --init --recursive
./install_executorch.sh
cd ../../..
```

--------------------------------

### Run LLAMA3.2 3B Instruct Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

This command executes the default example for LLAMA3.2 3B Instruct using hybrid mode.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --checkpoint consolidated.00.pth --params params.json --tokenizer_model tokenizer.model --decoder_model llama3_2-3b_instruct --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Install Llama Specific Requirements

Source: https://github.com/pytorch/executorch/blob/main/examples/models/deepseek-r1-distill-llama-8B/README.md

Installs the necessary requirements for running Llama models.

```bash
./examples/models/llama/install_requirements.sh
```

--------------------------------

### Run Codegen2 Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

This command executes the default example for Codegen2 using kv mode.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --decoder_model codegen2_1b --model_mode kv --max_seq_len 1024 --prompt "def hello_world():"
```

--------------------------------

### Example Usage

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llm_utils/README.md

This command demonstrates how to run the evaluation script with essential parameters.

```bash
python examples/qualcomm/oss_scripts/llm_utils/eval_decoder_model_qnn.py \
    --artifact ./eval_output \
    --tokenizer_path /path/to/your/tokenizer.model \
    --pte /path/to/your/model.pte \
    --soc_model SM8550 \
    --device YOUR_DEVICE_ID \
    --host localhost \
    --tasks wikitext \
    --limit 1 \
    --max_seq_len 512
```

--------------------------------

### Building and installing ExecuTorch with MPS support

Source: https://github.com/pytorch/executorch/blob/main/examples/apple/mps/README.md

Commands to build and install ExecuTorch, enabling MPS support.

```bash
cmake -DCMAKE_INSTALL_PREFIX=cmake-out \
          -DCMAKE_BUILD_TYPE=Release \
          -DEXECUTORCH_BUILD_DEVTOOLS=ON \
          -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \
          -DEXECUTORCH_BUILD_MPS=ON \
          -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
          -Bcmake-out .
cmake --build cmake-out -j9 --target install --config Release
```

--------------------------------

### Install specific transformers commit

Source: https://github.com/pytorch/executorch/blob/main/backends/mlx/examples/llm/README.md

Installs a specific commit of the transformers library, as required for Gemma 4 validation.

```bash
pip install -U "transformers @ git+https://github.com/huggingface/transformers.git@61461a7bcb458db7cf6eeea49678b9ab776a7821"
```

--------------------------------

### Exporting and Lowering to Ethos-U Backend

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/arm-ethos-u/tutorials/ethos-u-getting-started.md

This code snippet demonstrates how to export a quantized PyTorch model, partition it for the Ethos-U backend, and lower it to an Edge program, finally converting it to an ExecuTorch program and saving it as a .pte file.

```python
quantized_exported_program = torch.export.export(quantized_graph_module, example_inputs)
from executorch.backends.arm.ethosu import EthosUPartitioner
from executorch.exir import (
    EdgeCompileConfig,
    ExecutorchBackendConfig,
    to_edge_transform_and_lower,
)
from executorch.extension.export_util.utils import save_pte_program

# Create partitioner from compile spec
partitioner = EthosUPartitioner(compile_spec)

# Lower the exported program to the Ethos-U backend
edge_program_manager = to_edge_transform_and_lower(
            quantized_exported_program,
            partitioner=[partitioner],
            compile_config=EdgeCompileConfig(
                _check_ir_validity=False,
            ),
        )

# Convert edge program to executorch
executorch_program_manager = edge_program_manager.to_executorch(
            config=ExecutorchBackendConfig(extract_delegate_segments=False)
        )


# Save pte file
save_pte_program(executorch_program_manager, "ethos_u_minimal_example.pte")
```

--------------------------------

### Setup necessary tools for Ethos-U55/85

Source: https://github.com/pytorch/executorch/blob/main/examples/arm/README.md

Initial setup script to install necessary tools for building applications for Ethos-U55/85.

```bash
# Step [1] - setup necessary tools
$ cd <EXECUTORCH-ROOT-FOLDER>
$ ./examples/arm/setup.sh --i-agree-to-the-contained-eula

```

--------------------------------

### Install Requirements

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/qaihub_scripts/stable_diffusion/README.md

Shell script to install necessary Python packages for running the Stable Diffusion example.

```bash
sh examples/qualcomm/qaihub_scripts/stable_diffusion/install_requirements.sh
```

--------------------------------

### Change directory to Qualcomm examples

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/README.md

Navigate to the scripts directory within the Qualcomm examples for ExecuTorch.

```bash
cd $EXECUTORCH_ROOT/examples/qualcomm/scripts
```

--------------------------------

### Run Gemma 2B Example

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md

This command executes the default example for Gemma 2B using hybrid mode.

```bash
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 --decoder_model gemma-2b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
```

--------------------------------

### Setup Script

Source: https://github.com/pytorch/executorch/blob/main/docs/source/backends/nxp/nxp-overview.md

Command to run the setup script for installing development requirements, excluding MCUXpresso IDE and SDK.

```bash
./examples/nxp/setup.sh
```

--------------------------------

### Build ExecuTorch with LLM preset

Source: https://github.com/pytorch/executorch/blob/main/examples/models/phi-3-mini/README.md

Configures and builds the ExecuTorch environment with the LLM preset.

```bash
cmake --workflow llm-release
```

--------------------------------

### Install Executorch with Vulkan Support

Source: https://github.com/pytorch/executorch/blob/main/examples/vulkan/README.md

Installs the executorch python package from source with Vulkan backend support enabled.

```shell
CMAKE_ARGS="-DEXECUTORCH_BUILD_VULKAN=ON " ./install_executorch.sh -e
```

--------------------------------

### Install Hexagon SDK and Tools

Source: https://github.com/pytorch/executorch/blob/main/examples/qualcomm/custom_op/README.md

Examples of using qpm-cli to install specific versions of the Hexagon SDK and tools for different targets.

```bash
# example: hexagon-sdk-5.4.0 for v75 target (bundled with Hexagon tools 8.7.03)
qpm-cli --install hexagonsdk5.x --version 5.4.0.3 --path /path/to/Qualcomm/Hexagon_SDK/hexagon-sdk-5.4.0
# example: hexagon-sdk-6.0.0 for x86 target
qpm-cli --install hexagonsdk6.x --version 6.0.0.2 --path /path/to/Qualcomm/Hexagon_SDK/hexagon-sdk-6.0.0
```

```bash
# example: tools 8.8.02 for x86 target
qpm-cli --extract hexagon8.8 --version 8.8.02.1 \
  --path /path/to/Qualcomm/Hexagon_SDK/hexagon-sdk-6.0.0/tools/HEXAGON_Tools/8.8.02
```