v2.8.0-based

• Use TensorFlow 2.8
• Add more internal patches

TensorFlow v2.6.0 based

Push some internal patches (#7)

* Support new APUWareUtils

There is new version of APUWareUtils in Android 12.

* Support scalar input on PReLU and LeakyReLU

* Support quantized LeakyReLU

* Handle version 2 of TransposeConv

Update the verification to version 2 for INT8 TransposeConv.

* Support single input Reshape

* Fix handling of int8 inputs in TransposeConv

* Fix int8 delegation for activations

* Check Concatenation input type

* Support TransposeConv with bias

* Support the following OPs

- FullyConnected with keep_num_dims == true
- Pack

* Reject Gather with non-constant positions tensor

* Relax the limitation of FullyConnectede without bias

* print unsupport op name instead of op code

Co-authored-by: Code Lin <[email protected]>

moving to TF v2.5.x

moving to TF v2.5.x

v2.4.1

for TF v2.4.1