Chunking for mixed states #1249
-
Hello! Just wondering are there any plans for implementing chunking for mixed states? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
It's indeed not implemented yet, but simply because nobody needed it yet. The forward pass is already chunked, so you only need to chunk the backward. Though it might not be trivial to do it. You can find inspiration by looking at how the standard |
Beta Was this translation helpful? Give feedback.
It's indeed not implemented yet, but simply because nobody needed it yet.
It should be relatively straightforward to get it done. You need to re-implement this function, that computes the gradient of
LdagL
to use chunked primitives.The forward pass is already chunked, so you only need to chunk the backward. Though it might not be trivial to do it.
You can find inspiration by looking at how the standard
expect_and_grad
for hermitian operators is implemented without chunking and with chunking