DirectML

DirectML is a high-performance, hardware-accelerated DirectX 12 library for machine learning. DirectML provides GPU acceleration for common machine learning tasks across a broad range of supported hardware and drivers, including all DirectX 12-capable GPUs from vendors such as AMD, Intel, NVIDIA, and Qualcomm.

MIT License

Downloads
40.5K
Stars
2.2K
Committers
46

Bot releases are hidden (Show)

DirectML - torch-directml 0.2.4.dev240815 Latest Release

Published by joshjkim 2 months ago

Preview build of torch-directml built on August 15, 2024.

The Python packages are available as a PyPI release. To download the appropriate python package automatically, simply pip install torch-directml.

What’s New

  • Operator support for avg_pool3d, upsample_bicubic2d
  • Support for device_count in torch PrivateUse1 backend

Bug Fixes

  • Fix for Whisper sample not running in fp16 with DirectML scaled dot product attention #598
  • Issues addressed with operator support and device_count inPrivateUse1 #596 #609
DirectML - torch-directml 0.2.3.dev240715

Published by joshjkim 3 months ago

Preview build of torch-directml built on July 15, 2024.

The Python packages are available as a PyPI release. To download the appropriate python package automatically, simply pip install torch-directml.

What’s New

  • Operator support for isin, std_mean, group_norm, multinomial
  • Added samples and support for Stable Diffusion Turbo and XL Turbo
DirectML - torch-directml 0.2.2.dev240614

Published by joshjkim 4 months ago

Preview build of torch-directml built on June 14, 2024.

The Python packages are available as a PyPI release. To download the appropriate python package automatically, simply pip install torch-directml.

What’s New

  • Support for PyTorch 2.3.1 and torchvision 0.18.1
  • Operator support for softplus, amax, linspace, vector_norm, native_dropout
  • Performance improvements for linear operator
  • Initial support for SparseTensor via dense tensors
  • Added sample and support for OpenAI Whisper model
  • LLM sample code refactor and cleanup for improved maintainability
  • Added Generator register and support for random seed for torch PrivateUse1 backend

Bug Fixes

  • Fix for layer norm producing incorrect result with non-contiguous input #588
  • Issues addressed with new operators and PrivateUse1 random seed support #592 #590 #587 #586
DirectML - tensorflow-directml-1.15.3.dev200626

Published by jstoecker over 4 years ago

Preview build of tensorflow-directml built on June 26, 2020.

The Python packages are available as a PyPI release. To download the appropriate python package automatically, simply pip install tensorflow-directml.

Changes in dev200626:

• Certain OOM errors will now gracefully fail instead of crashing the python interpreter.
• Allow usage of entire graphics memory when grappler optimizations are enabled.
• GRUBlockCell operator implemented.

DirectML - tensorflow-directml-1.15.3.dev200615

Published by jstoecker over 4 years ago

First preview of TensorFlow 1.15.3 with DirectML support in both Windows and WSL.

Please visit tensorflow-directml on PyPI.org for the latest release!