Open shivadbhavsar opened 1 month ago
MIGraphX does not yet have a GPU fill operator. Would need to implement it to complete this issue.
We might be able to partially resolve this in cases where the fill input is a constant. There are some passes that do this for torch.zeros, torch.full, torch.arange
etc
Add aten converter. Affected models: Torchbench: pyhpc_turbulent_kinetic_energy