Hieu Pham
[Google Scholar]
[GitHub]
[Email]
[Personal Blog]
I am an engineer at
OpenAI.
These days, I am interested in writing fast kernels for neural networks.
This process is very time consuming, so I also try to come up with ways to help kernel developers complete their jobs faster.
When it comes to programming for NVIDIA GPUs, I highly recommend CUTLASS, especially the Python Cute DSL for development speed.
When it comes to generic programming on chips, I highly recommend Triton and Gluon.
I believe a neural compiler that allows programmers to write fast kernels in a Python interface is the future.
And I am working hard to make it happen.