v1.4.0 comes with two major changes:
- CLI tools for benchmarking are now available in the package:
- You can now run
python -m neutone_sdk.benchmark benchmark-speed --model_file <path-to-model.nm>orpython -m neutone_sdk.benchmark benchmark-latency --model_file <path-to-model.nm>for speed and latency benchmarks respectively. - Profiling is also possible with
python -m neutone_sdk.benchmark profile --model_file <path-to-model.nm>. - Check the dedicated README section for more detailed information.
- A set of default benchmarks are also ran automatically when a model is exported.
- You can now run
- The limit on PyTorch is lifted to 2.2.0, enabling users to use the newly released 2.1.0.