tree ddbec4e57335e558acd18bd94278177b24621e0c
parent 649966f38f533a7e74913d8edf2989325a186a38
author Robert David <lrdx@google.com> 1605810921 -0800
committer TensorFlower Gardener <gardener@tensorflow.org> 1605811627 -0800

Bump FULLY_CONNECTED's supported version to v9.

New options and features introduced between v4 and v9:
v5: keep_num_dims: Not supported. Added check to FullyConnectedOperationParser to explicitly fail when flag is set.
v6: No bias provided: Was already supported in GPU, but the versioning not allowed to run these models.
v7: INT16 input/output. Integer input/output tensors are not supported.
v8: sparse_weight: Sparsity is not supported. Added check to ObjectReader::ReadTensor to reject sparse tensors.
v9: asymmetric_quantize_inputs: Controls on-the-fly quantization behavior with hybrid quant. Not relevant for GPU, as all operations are in floating-point.
PiperOrigin-RevId: 343320830
Change-Id: I4198173a83f1063099df18f531c478208f3594a6