This release of AMD's CPU solution for TensorFlow provides a binary built with the PluggableDevice approach.
zentf v5.1 Release Highlights
- Added support for TensorFlow 2.19.
- Java® Integration: Enabled support for PluggableDevice in TensorFlow-Java v1.0.0.
- Upstreamed the Java Integration to the TensorFlow-Java repository, strengthening its core capabilities.
- Integrated with ZenDNN v5.1 as the core inference library and is compiled and tested with GCC v13.3.
- Introduced new operator fusions to accelerate common computation patterns:
- MatMul + BiasAdd + Sigmoid
- MatMul + BiasAdd + Tanh fusions
Highlights of the Last Major Release (zentf 5.0)
- Merged BF16 and FP32 compute flows and added broadcasting support for BatchMatMul kernel.
- INT8 support for the ResNet50 model.
- Softmax kernel supports up to 5D.
- Deprecated blocked format support for convolution ops and restriction of rewrite for the fused ops based on the post ops.
- Provided experimental support of C++ APIs.