DefaultCPUAllocator: can't allocate memory

[LylianChallier]

Hello !

I have trained a [15, [1, 1], 2] KAN model on 2D Rayleigh-Benard DNS data. When I use model.fix_symbolic(), then I get a memory error : " RuntimeError: [enforce fail at alloc_cpu.cpp:118] err == 0. DefaultCPUAllocator: can't allocate memory: you tried to allocate 102112010000 bytes. Error code 12 (Cannot allocate memory) ".

Is the fix_symbolic method very expensive in memory ? How does it work ?

Thanks a lot !

[KindXiaoming]

It is not particularly a big KAN, so I guess you are using a very large batch size. You could try reducing the batch size.