To further expand our support on channels last, we want to further expand this to arbitrary permutation.
The challenge would be to maintain a coherent behavior as with eager (TensorIterator). Might make sense to refactor eager code so the logic of permutation extraction could be re-used in nvfuser integration.
jjsjann123
commented
2 years ago
🚀 Feature
To further expand our support on channels last, we want to further expand this to arbitrary permutation. The challenge would be to maintain a coherent behavior as with eager (TensorIterator). Might make sense to refactor eager code so the logic of permutation extraction could be re-used in nvfuser integration.