SVD gradient has not been implemented for input with unknown inner matrix shape.
grad_v_shape[:-2]) a_shape = batch_shape.concatenate([m, n]) m = a_shape.dims[-2].value n = a_shape.dims[-1].value # TODO(rmlarsen): Make this work with placeholders. if m is None or n is None: raise NotImplementedError( "SVD gradient has not been implemented for input with unknown " "inner matrix shape.") s = op.outputs u = op.outputs v = op.outputs s = math_ops.cast(s, a.dtype) u
NO FIXES YET
Just press the button and we will add solution
to this exception as soon as possible
* As many users press the button, the faster we create a fix
Add a possible fix
Pleaseto post fix