The latest release of uxlfoundation/onednn is v3.13.4. oneDNN (oneAPI Deep Neural Network Library) is an open source library that provides deep learning operations such as matrix multiplication, convolution, and attention on CPUs and Intel GPUs. The provided information does not separately list what changed in this version.
This library is optimized for x64, AArch64, RISC-V CPUs and Intel GPUs. The README introduces it as a cross-platform deep learning compute library. It provides optimized implementations of matrix multiplication, convolution, attention, and normalization, and is distributed under the Apache-2.0 license. PyTorch, TensorFlow, OpenVINO, and ONNX Runtime use oneDNN as an acceleration layer. So rather than replacing frameworks, oneDNN is closer to the foundation that implements the deep learning operations those frameworks call.
The provided information contains no specific changes for this release and no performance measurements. Therefore it is not possible to judge here how v3.13.4 changed functionality or speed. What can be confirmed is only that it targets multiple CPU families and Intel GPUs, and that it is a compute library used by major frameworks.