Help needed. Is there a way to call the function inside a numba @cuda.jit kernel function? if not how should i use the package on a GPU?

[anbonimus]

Hi I was trying to use the odeint funciton within a numba cuda jit kernel function, but i got the error msg

numba.core.errors.TypingError: Failed in cuda mode pipeline (step: nopython frontend)
Untyped global name 'odeint': Cannot determine Numba type of <class 'function'>

File "temp.py", line 138:
def ode_kernel(Tk_avg):

t = torch.linspace(0,tf,2000)#200000
y = odeint(atom,torch.Tensor(ini_state),t,rtol=10**-10,atol=10**-20,method='scipy_solver',options={'solver':'Radau'})
^

During: Pass nopython_type_inference

how should i make it work? If torchdiffeq is not compatible with numba, how should i use it on GPU?

Thanks in advance!