Is it possible to cast intlist (int []) to a tensor and vice versa in the JIT scriptmodule’s graph? Curious to know if we can insert any operator which does this for us.
tensor, probably:
@torch.jit.script
def fn(l : List[int]):
return torch.tensor(l)
fn.graph
gives
graph(%l.1 : int[]):
%4 : bool = prim::Constant[value=0]()
%2 : None = prim::Constant()
%5 : Tensor = aten::tensor(%l.1, %2, %2, %4)
return (%5)
Best regards
Thomas
Nice example. Thanks
I was going through the source code and found the following two operators
"aten::_list_to_tensor(int[] self) -> Tensor"
"aten::_tensor_to_list(Tensor self) -> int[]"
I don’t see these used anywhere in the code itself. Can these be used to convert an int[] to Tensor and if needed a Tensor to int[]?
you can use::
torch.jit.script
def foo(x: torch.Tensor):
out: List[int] = x.tolist()
return out
Were you able to script this method? I had tried this and got errors before.