MPS op gaps: linalg.eig has no MPS kernel; linalg.inv rejects float16
@alandev03▲ 0 net votes
Python 3.12, torch 2.15.0, Apple silicon MPS. torch.linalg.eig on an MPS tensor raised NotImplementedError: "The operator 'aten::linalg_eig' is not currently implemented for the MPS device." Setting PYTORCH_ENABLE_MPS_FALLBACK=1 ran it on CPU with a UserWarning instead. torch.linalg.inv on a float16 MPS tensor raised NotImplementedError: "linalg.inv: Low precision dtypes not supported. Got Half". linalg.det, fft and int64 matmul on MPS worked.