Hi,
I wanted to try using the matmul op in the conv1d implementation (i.e. task 4.1), but the tests are being run on the simple backend, the only one of the three which lacks a matmul implementation. I find this a bit weird especially because (1) conv1d is supposed to be implemented in fast_conv.py which made me expect the tests to use the fast backend, and (2) the guide on convolution actually makes use of the matmul op itself.
Should students essentially ignore both previous matmul implementations when working on conv1d and write it more from scratch with multiple nested fors?
Edit: It wasn't that hard to do it from scratch after all, just a bit confusing given the above
Hi,
I wanted to try using the
matmulop in theconv1dimplementation (i.e. task 4.1), but the tests are being run on the simple backend, the only one of the three which lacks amatmulimplementation. I find this a bit weird especially because (1)conv1dis supposed to be implemented infast_conv.pywhich made me expect the tests to use thefastbackend, and (2) the guide on convolution actually makes use of thematmulop itself.Should students essentially ignore both previous
matmulimplementations when working onconv1dand write it more from scratch with multiple nested fors?Edit: It wasn't that hard to do it from scratch after all, just a bit confusing given the above