RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 5 but got size 10 for tensor number 1 in the list. about torchdeq HOT 2 CLOSED

Hi @georgewanglz2019 ,

Thank you for your interest in TorchDEQ! I apologize for my late response. (Busy working on the Consistency Models Made Easy blog post. Recommend it also. :)

Yeah. Indeed, you should pass a Python functor that only takes the equilibrium tensor z as input into the DEQ module. The DEQ module will solve for the fixed point z* and track its gradient.

If you have a joint equilibrium formulation, for example, [a*, b*, c*] = f([a*, b*, c*], x), you can also code it like this.

Let me know if you have further questions!

Thanks,
Zhengyang

from torchdeq.

modified:

x_, z_ = torch.ones(5), torch.zeros(10) # z0
f = lambda z: fcnn(x_, z)
z_out, info = deq(f, z_)

maybe because I should define a function f which only have one input "z", now it works!

from torchdeq.