is the pytorch backend differentiable?

Question

is the pytorch backend differentiable?

ni-chen opened this issue 2 years ago · comments

Ni Chen commented 2 years ago

Hi,

Thanks for making this amazing package, this could be useful.
May I ask if the PyTorch backend is differentiable?

Thanks.

Floris Laporte · Answer 1 · Tue Apr 19 2022 02:50:57 GMT+0800 (China Standard Time)

Hey @ni-chen ,

this has been on my todo list for too long, currently there are a few in-place operations preventing this... However, I don't think it should be too difficult to change, I expect the simulator to work slower, however...

Maybe when I find the time I might finally give it a shot soon. This would obviously be an awesome feature to have.

Ni Chen · Answer 2 · Tue Apr 19 2022 15:50:03 GMT+0800 (China Standard Time)

Thanks for your reply.
I believe it would be of great interest for many applications. @flaport

simenhu · Answer 3 · Tue Sep 26 2023 22:20:36 GMT+0800 (China Standard Time)

Would you be able to point to the places where the changes would have to be made? Could see if I could try to make a PR on it if it's not a very big change?

Would the general approach of preallocating the field tensors and write to a new index instead of updating an already populated tensor be the general solution you are thinking of?