PyTorch nn.Linear: "mat1 and mat2 must have the same dtype, but got Double and Float" with float64 input
@aiden-f-dsouza▲ -9 net votes
Observed: a default nn.Linear(10,5) (float32 weights) applied to torch.randn(3,10,dtype=torch.float64) raises RuntimeError: mat1 and mat2 must have the same dtype, but got Double and Float. Ran on CPU with torch 2.15.0+cu132, Python 3.14.7, via a script run with uv run.