Alexey Chernov

@mohit-bansal Try not to use *`:cpu`* postfix for the container name as you're probably building it from the plain `Dockerfile`, i.e. try as `sudo docker run -v $PWD:/opt/app -e PYTHONPATH=$PYTHONPATH:/opt/app...

@mohit-bansal you're right, if I remember correctly, training takes quite long on the CPU. I don't know the exact number of epochs to have the model converged enough, I made...

Thank you for clarification on the self test. As for second, yes, I guess, there's some incorrectness in the initialization of using FIFO, but can't find, what exactly. If I...

To keep it resolved here, in case of someone would search for similar problem, in my case the issue was with wrong implementation of writing operation to MPU register through...

That's probably a way to go, but the problem is `ncc` refuses to parse valid ONNX passed from vanilla framework with some strange error, while it has all the necessary...

Sorry for delay with the test failure fixes I promised to make, was a little busy this time. I seem to fix the failure putting the output shape generation code...

@iotamudelta thanks for your attention and suggestions! Yes, I'll file a question there as well with the reference here. As for the building, I pretty much replicate the building process...

Yes, in case of ROCm build I both set `USE_ROCM=1` and perform `tools/amd_build/build_amd.py` script to do hipification. It builds quite fine, but I'm not sure about passing `USE_ROCM=1` to `setup.py`...

Can't confirm it for 5.0, haven't built PyTorch against it yet, unfortunately. Feel free to close it for now, I'll reopen if it's still there.

No problem) Yes, I've noticed the package seems to be completely rewritten now. Many thanks for your kind assistance with many of my previous requests!