ExecuTorch-rs

executorch is a Rust library for executing PyTorch models in Rust. It is a Rust wrapper around the ExecuTorch C++ API. It depends on version 0.5.0 of the Cpp API, but will advance as the API does. The underlying C++ library is still in Beta, and its API is subject to change together with the Rust API.

Usage

Create a model in Python and export it:

import torch
from executorch.exir import to_edge
from torch.export import export

class Add(torch.nn.Module):
    def __init__(self):
        super(Add, self).__init__()

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


aten_dialect = export(Add(), (torch.ones(1), torch.ones(1)))
edge_program = to_edge(aten_dialect)
executorch_program = edge_program.to_executorch()
with open("model.pte", "wb") as file:
    file.write(executorch_program.buffer)

Execute the model in Rust:

use executorch::evalue::IntoEValue;
use executorch::module::Module;
use executorch::tensor_ptr;
use ndarray::array;

let mut module = Module::new("model.pte", None);

let (tensor1, tensor2) = (tensor_ptr![1.0_f32], tensor_ptr![1.0_f32]);
let inputs = [tensor1.into_evalue(), tensor2.into_evalue()];

let outputs = module.forward(&inputs).unwrap();
assert_eq!(outputs.len(), 1);
let output = outputs.into_iter().next().unwrap();
let output = output.as_tensor().into_typed::<f32>();

println!("Output tensor computed: {:?}", output);
assert_eq!(array![2.0], output.as_array());

See example/hello_world for a complete example.

Build

To build the library, you need to build the C++ library first. The C++ library allow for great flexibility with many flags, customizing which modules, kernels, and extensions are built. Multiple static libraries are built, and the Rust library links to them. In the following example we build the C++ library with the necessary flags to run example hello_world:

# Clone the C++ library
cd ${EXECUTORCH_CPP_DIR}
git clone --depth 1 --branch v0.5.0 https://github.com/pytorch/executorch.git .
git submodule sync --recursive
git submodule update --init --recursive

# Install requirements
./install_requirements.sh

# Build C++ library
mkdir cmake-out && cd cmake-out
cmake \
    -DDEXECUTORCH_SELECT_OPS_LIST=aten::add.out \
    -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=OFF \
    -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=OFF \
    -DBUILD_EXECUTORCH_PORTABLE_OPS=ON \
    -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
    -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
    -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
    -DEXECUTORCH_ENABLE_PROGRAM_VERIFICATION=ON \
    -DEXECUTORCH_ENABLE_LOGGING=ON \
    ..
make -j

# Static libraries are in cmake-out/
# core:
#   cmake-out/libexecutorch.a
#   cmake-out/libexecutorch_core.a
# kernels implementations:
#   cmake-out/kernels/portable/libportable_ops_lib.a
#   cmake-out/kernels/portable/libportable_kernels.a
# extension data loader, enabled with EXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON:
#   cmake-out/extension/data_loader/libextension_data_loader.a
# extension module, enabled with EXECUTORCH_BUILD_EXTENSION_MODULE=ON:
#   cmake-out/extension/module/libextension_module_static.a
# extension tensor, enabled with EXECUTORCH_BUILD_EXTENSION_TENSOR=ON:
#   cmake-out/extension/tensor/libextension_tensor.a
# extension tensor, enabled with EXECUTORCH_BUILD_DEVTOOLS=ON:
#   cmake-out/devtools/libetdump.a

# Run example
# We set EXECUTORCH_RS_EXECUTORCH_LIB_DIR to the path of the C++ build output
cd ${EXECUTORCH_RS_DIR}/examples/hello_world
python export_model.py
EXECUTORCH_RS_EXECUTORCH_LIB_DIR=${EXECUTORCH_CPP_DIR}/cmake-out cargo run

The executorch crate will always look for the following static libraries:

• libexecutorch.a
• libexecutorch_core.a

Additional libs are required if feature flags are enabled (see next section):

• libextension_data_loader.a
• libextension_module_static.a
• libextension_tensor.a
• libetdump.a

The static libraries of the kernels implementations are required only if your model uses them, and they should be linked manually by the binary that uses the executorch crate. For example, the hello_world example uses a model with a single addition operation, so it compile the C++ library with DEXECUTORCH_SELECT_OPS_LIST=aten::add.out and contain the following lines in its build.rs:

println!("cargo::rustc-link-lib=static:+whole-archive=portable_kernels");
println!("cargo::rustc-link-lib=static:+whole-archive=portable_ops_lib");

let libs_dir = std::env::var("EXECUTORCH_RS_EXECUTORCH_LIB_DIR").unwrap();
println!("cargo::rustc-link-search=native={libs_dir}/kernels/portable/");

Note that the ops and kernels libs are linked with +whole-archive to ensure that all symbols are included in the binary.

The build (and library) is tested on Ubuntu and MacOS, not on Windows.

Cargo Features

• data-loader

  Includes the FileDataLoader and MmapDataLoader structs. Without this feature the only available data loader is BufferDataLoader. The libextension_data_loader.a static library is required, compile C++ executorch with EXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON.

• module

  Includes the Module struct. The libextension_module_static.a static library is required, compile C++ executorch with EXECUTORCH_BUILD_EXTENSION_MODULE=ON. Also includes the std feature.

• tensor-ptr

  Includes the TensorPtr struct, a smart pointer for tensors that manage the lifetime of the tensor object alongside the lifetimes of the data buffer and additional metadata. The libextension_tensor.a static library is required, compile C++ executorch with EXECUTORCH_BUILD_EXTENSION_TENSOR=ON. Also includes the std feature.

• etdump

  Includes the ETDumpGen struct, an implementation of an EventTracer, used for debugging and profiling. The libetdump.a static library is required, compile C++ executorch with EXECUTORCH_BUILD_DEVTOOLS=ON and EXECUTORCH_ENABLE_EVENT_TRACER=ON. In addition, the flatcc (or flatcc_d) library is required, available at {CPP_EXECUTORCH_DIR}/third-party/flatcc/lib/, and should be linked by the user.

• ndarray

  Conversions between executorch tensors and ndarray arrays. Adds a dependency to the ndarray crate. This feature is enabled by default.

• half

  Adds a dependency to the half crate, which provides a fully capable f16 and bf16 types. Without this feature enabled, both of these types are available with a simple conversions to/from u16 only. Note that this only affect input/output tensors, the internal computations always have the capability to operate on such scalars.

• num-complex

  Adds a dependency to the num-complex crate, which provides a fully capable complex number type. Without this feature enabled, complex numbers are available as a simple struct with two public fields without any operations. Note that this only affect input/output tensors, the internal computations always have the capability to operate on such scalars.

• std

  Enable the standard library. This feature is enabled by default, but can be disabled to build executorch in a no_std environment. See the examples/no_std example. Also includes the alloc feature. NOTE: no_std is still WIP, see https://github.com/pytorch/executorch/issues/4561

• alloc

  Enable allocations. When this feature is disabled, all methods that require allocations will not be compiled. This feature is enabled by the std feature, which is enabled by default. Its possible to enable this feature without the std feature, and the allocations will be done using the alloc crate, that requires a global allocator to be set.

By default the std and ndarray features are enabled.