This end-to-end example shows how to:
Resnet 18
with tvm
from PythonTo run the example with pretrained resnet weights, first tvm
and mxnet
must be installed for the python build. To install mxnet for cpu, run pip install mxnet
and to install tvm
with llvm
follow the TVM installation guide.
To have a successful build, note that it is required to instruct Rust compiler to link to the compiled shared library, for example with println!("cargo:rustc-link-search=native={}", build_path)
. See the build.rs
for more details.
cargo run
Note: To use pretrained weights, one can enable --pretrained
in build.rs
with
let output = Command::new("python") .arg(concat!(env!("CARGO_MANIFEST_DIR"), "/src/build_resnet.py")) .arg(&format!("--build-dir={}", env!("CARGO_MANIFEST_DIR"))) .arg(&format!("--pretrained")) .output() .expect("Failed to execute command");
Otherwise, random weights are used, therefore, the prediction will be limpkin, Aramus pictus
!