What's new in WML CE 1.7.0
Check here for the latest updates to IBM Watson Machine Learning Community Edition information for the 1.7.0 release.
What's new
- pai4sk and snapml-spark conda packages are now available on x86. See Getting started with IBM Distributed Accelerated ML library.
- snapml-spark now supports Spark 2.4 in addition to Spark 2.3. See Getting started with Snap Machine Learning (SnapML) on Apache Spark.
- Multi-threaded CPU training of DecisionTree algorithms in pai4sk.
- GPU-accelerated training of DecisionTree and RandomForest algorithms in pai4sk. See Getting started with IBM Distributed Accelerated ML library.
- RandomForest supports multi-GPU acceleration.
- Applications using pai4sk APIs can use up to two GPUs from a single node without a IBM Watson® Machine Learning Accelerator.
- Support for the IBM Power® Systems IC922 and the NVIDIA T4 Tensor Core GPU
- CUDA 10.2 Support
- TensorFlow 2.1 with eager execution and the redesigned APIs for TensorFlow 2
- PyTorch 1.3.1
- Horovod 0.19. See Getting started with Horovod for details.
Technology previews
- pai4sk: SnapBoost algorithm - see Getting started with IBM Distributed Accelerated ML library.
- Dask support for GPU-backed dataframe (dask-cudf) and multi-GPU machine learning algorithms - see cuML and XGBoost.
- CuPy: NumPy-like API accelerated with CUDA - see cuPy in Getting started with RAPIDS.
- LMS for TensorFlow 2. See Getting started with TensorFlow large model support
Note: For licence information for technology previews, see Technology Preview Code.
Deprecations
- The DDL Tensorflow package and DDL PyTorch backend will be deprecated in future releases. It is recommended to use the new Horovod packages for distributed deep learning.