Connor Clark

I would strongly discourage using the low level interface directly. It exposes implementation details of the core gRPC library that you don't get access to in other languages' bindings (so...

To clarify this, in my tests previously, calling [`grpc_call_cancel`](http://www.grpc.io/grpc/core/grpc_8h.html#ada3c794f1c02477b0d71b12026dd0333) on the server side after a streaming call has started doesn't have any effect. I wasn't able to find any place...

Looking at the types again, I agree with @ixmatus that the type seems to be accounting for status codes that don't include a response message, and all of our high...

Same error with [the latest Nvidia pytorch Docker image](https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-05.html#rel-23-05). It happens to me with both `0.9.0` and `0.9.4` (and presumably every version in between). Nvidia driver version: 525.116.03. Using `nvidia-container-runtime`....

Setting `NVCC_PREPEND_FLAGS="--forward-unknown-opts"` appears to be a workaround for this issue.

Thanks for filing the issue! As always, the tedious part is updating the CI. I have a very messy WIP branch with a draft PR. I'm hitting some odd behavior...

This is indeed confusing. I just tried it with the official Python implementation, and both cases result in `foo = 1`. I believe the key is interpreting the phrase >...

More concrete docs for this are available [here](https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/examples/advanced_optimizations.html#Gradient-accumulation-fusion).