I’ve started using JIT tracing with a RNN I’ve written. By following the seq2seq tutorial, the CPU inference speed has improved by 3x. Is this the expected speed improvement? If I were to compile the module into C++, would I see further improvements?
1 Like