microsoft / microsoft/onnxruntime
Explain in the docs that run() or run_with_binding() always synchronizes the CUDA stream prior to returning control (or an explicit note about proper benchmarking ONNX ops versus PyTorch ops also wrt CUDA syncs)
Open
@natke is already working on this.
Since Oct 21, 2021.
documentation
- Dominant language
- C++
- Stars
- 21.9k
- Forks
- 4.2k
- Avg merge
- 4d 8h
- Merged PRs (30d)
- 179
Description
It is important for simplifying benchmarks as in https://github.com/microsoft/onnxruntime/issues/7212, i.e. avoid GPU->CPU transfer just to ensure the model actually completed execution.
I found that execution providers have method Sync(). How to call it from Python?
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.