Error on Apple M1 pro

[ChenNed]

KeyError: 'mps'

When I tried to use DGLGraph.to('mps'), a "keyerror" was issued.

Data error when use tensorflow

When I tried to load the Cora dataset, the following error occurred:

“_dgl.ffi.base.DGLError: Cannot assign node feature "train_mask" on device /gpu:0 to a graph on device /cpu:0. Call DGLGraph.to() to copy the graph to the same device.”

Note: I install the latest version of dgl (0.9.0), pytorch (1.12.0) and tensorflow-metal(2.9).

How can I solve this problem.

[BarclayII]

For the first one, DGL currently does not support mps device on M1. The second one looks like a bug on tensorflow support.

[jermainewang]

Does 'mps' device require special kernel implementation?