Skip to content

zentf Release v5.2

Latest
Latest

Choose a tag to compare

View all tags
@kiriti-pendyala kiriti-pendyala released this 12 Mar 16:56
v5.2
2b82f4b

ZenDNN Plugin for TensorFlow (zentf) — Release Notes v5.2

Overview

zentf 5.2 is a major release that continues our focus on optimizing inference for Recommender Systems and Large Language Models on AMD EPYC™ CPUs.

What's New in zentf 5.2

TensorFlow Version Support

Component Details
TensorFlow 2.20.0 Primary supported version with optimal performance. Distributed as a Python wheel via PyPI and as a C++ package.
TensorFlow-Java main (75402bef) Java User Interface — Fully supported (available via source build only).

Improvements

1. TF 2.20 Integration

  • zentf 5.2 is built for and validated against TensorFlow v2.20.0.
  • Bazel 7.4.1 — Upgraded from the Bazel 5.3–6.5 range to a single supported version (7.4.1).
  • Because TensorFlow-Java is not released for TensorFlow 2.20.0, zentf supports TensorFlow-Java main (75402bef) via source build only.

2. Migrate from Legacy ZenDNN Library to ZenDNNL

  • CMake-based ZenDNNL integration using rules_foreign_cc.
  • All operator kernels (MatMul, Conv2D, BatchMatMul, Softmax, Pooling) have been rewritten to use the ZenDNNL Low Overhead API (LOA), replacing the legacy ZenDNN primitives.
  • Old third-party dependencies on zen_dnn and amd_blis (BLIS) have been removed, replaced by ZenDNNL with integrated AOCL-DLP.

3. Removed Legacy Components

  • Mempool optimization has been completely removed; equivalent performance is achieved using jemalloc as the memory allocator instead.
  • INT8 support has been removed.
  • Non-performant ops removed — ZenTranspose, ZenReshape, Binary ops.

4. Performance Optimizations

  • Enhanced Operations with LOA: Low Overhead API optimizations for improved performance.

Breaking Changes

Caution

  • Dropped TensorFlow Backward Compatibility: Backward compatibility with previous TensorFlow versions has been discontinued due to major changes in TensorFlow 2.20.0.
  • Removed Mempool Support: Dropped support for mempool optimization.
  • Dropped INT8 Support: Previously available only for the ResNet50 model; now fully removed.
  • Removed Ops: Cleaned up non-performant ops and obsolete fusions — ZenTranspose, ZenReshape, Binary ops, BatchNorm fusions.
Assets 2
Loading