Please consider also adding a small note to help those few that get stuck with this bug:
RuntimeError: expected scalar type BFloat16 but found Float
append the execution command with `--precision full`