|
|
|
|
|
|
by bjourne
284 days ago
|
|
|
Well, Mojo hasn't been released so we can't precisely say what it can and can't do. If it can emit CUDA code then it does it by transpiling Mojo into CUDA. And there is no reason why Python can't also be transpiled to CUDA. What I mean here is that DNN code is written on a much higher level than kernels. They are just building blocks you use to instantiate your dataflow. |
|
|