Certain softmax shapes won't convert to tensorflow lite

General Discussion
,
#1

Quick question. Why is it that certain softmax shapes will not convert to tensorflow lite? For example (10,26,26) will work but (10,26,26,1) will not. I am getting the following error

Some ops are not supported by the native TFLite runtime, you can enable TF kernels fallback using TF Select. See instructions: https://www.tensorflow.org/lite/guide/ops_select
TF Select ops: Softmax
Details:
	tf.Softmax(tensor<?x10x26x26x1xf32>) -> (tensor<?x10x26x26x1xf32>) : {device = ""}

I am trying to figure out how to solve this while keeping the tf runtime.

I also have a stack overflow topic

#3

Have you checked if is it still related to: