Repository navigation
Tensor.to(Tensor) doesn't change dtype of Tensor. #653
Description
PyTorch outputs torch.float64
import torch
x = torch.linspace(0, 1, 100, dtype=torch.float32)
y = torch.linspace(0, 1, 100, dtype=torch.float64)
x.to(y).dtypeTorchSharp (8005af9) outputs Float32.
// commit 8005af93ef3ebf66f3b246a4ccbe4ba791f7f9f9
using TorchSharp;
var x = torch.linspace(0, 1, 100, dtype: torch.float32);
var y = torch.linspace(0, 1, 100, dtype: torch.float64);
x.to(y).dtypeMetadata
Metadata
Assignees
Labels
No labels
Activity
kaiidams commentedon Jul 12, 2022
TorchSharp provides
Tensor.type_as(Tensor)to changeTensor.dtype, however, this difference from PyTorch is confusing when poirting from PyTorch. Also in many case,to()is used to apply operators to two tensors wheredtypeanddeviceof them need to be compatible.NiklasGustafsson commentedon Jul 20, 2022
@kaiidams -- this will be fixed in the next release (which I will release today).
Fixed bug dotnet#653
Merge pull request #659 from NiklasGustafsson/main