You should compile the pytorch code… otherwise I think you’re leaving a lot of performance on the table. See the docs for how to do that
You should compile the pytorch code… otherwise I think you’re leaving a lot of performance on the table. See the docs for how to do that