Is it possible to get a speedup in coreml by pruning the model such that the tensors are sparse? I know that tflite is capable of optimizing ops with sparse tensors. Wondering if this is a feature of coreml yet.
Does model pruning lead to speedup in coreml?
According to CoreMLTools documentation if you convert to ML Program format you could use sparse representation - https://coremltools.readme.io/docs/compressing-ml-program-weights
It should be interesting to know if this is supported on both NE, GPU and CPU.