Block Sparsity not working for Squeezable Pointwise Convolution Layers

[s36srini]

ValueError: Block Sparsity can only be used for layers which have 2-dimensional weights. I checked the source code and saw this comment:
# TODO(pulkitb): Check if squeeze operations should now be removed since we are only accepting 2-D weights.

How is pruning going to work with pointwise convolutional layers of format 1x1xiCxoC if block sparsity is only supported for 2D tensors?

[teijeong]

Hi @liyunlu0618 , can you update?

[liyunlu0618]

Duplicate of #634.
Will use that to track the feature request.