Add intel-extension-for-openxla.md #31
Conversation
wangkl2
commented
Jul 28, 2023
wangkl2
commented
- Add md for intel-extension-for-openxla blog
louie-tsai
left a comment
louie-tsai
left a comment
There was a problem hiding this comment.
looks good to me. just some minor inputs. possible to response those comments?
|
|
||
| OpenXLA is an open-source Machine Learning compiler ecosystem co-developed by AI/ML industry leaders that lets developers compile and optimize models from all popular ML frameworks on a wide variety of hardware. We are pleased to announce Intel® Extension for OpenXLA\*, which seamlessly runs AI/ML models on Intel GPUs. Intel® Extension for OpenXLA\* is a high-performance deep learning extension implementing the [OpenXLA](https://github.com/openxla/xla) PJRT C API (see [RFC](https://github.com/openxla/community/blob/main/rfcs/20230123-pjrt-plugin.md) for details), allowing multiple AI frameworks to compile StableHLO, an operation set for high-level operations for numeric computation, as well as dispatching the executable to Intel GPUs. | ||
|
|
||
| <br><div align=center><img src="/content/dam/developer/articles/technical/accelerate-stable-diffusion-on-intel-gpus-with-intel-extension-for-openxla/Intel-Data-Center-GPU-Max-Series-1100-PCIe-Card.png"></div> |
There was a problem hiding this comment.
one quick question, have the PNG file been in the AEM system already?
Is it for a new AEM/IDZ article or for an existed IDZ article?
There was a problem hiding this comment.
Yes, those images have been uploaded to the AEM system. The article has been published but hope to keep it in this repo for unified management for future modification.
|
|
||
| In this section, we show how to enable JAX applications on Intel GPUs with Intel® Extension for OpenXLA\*. | ||
|
|
||
| ### Installation |
There was a problem hiding this comment.
possible also put a reference link to official Github repo page or github.io doc if any?
| ### Install Required Packages | ||
| ```bash | ||
| pip install flax==0.6.6 transformers==4.27.4 diffusers==0.16.1 datasets==2.12.0 | ||
| sudo apt install numactl |
There was a problem hiding this comment.
looks like that we need Ubuntu.
Should we mention OS requirement in Preinstallation Requirement session?
|
|
||
| Setup environment variable to utilize 1 stack of the GPU (only for 2-stacks GPU, such as Max 1550 GPU) and use `numactl` to bind the process with GPU affinity NUMA node. | ||
| ```bash | ||
| export ZE_AFFINITY_MASK=0.0 |
There was a problem hiding this comment.
maybe good to put official Affinity mask doc below for a reference.
https://spec.oneapi.io/level-zero/0.91/core/INTRO.html#affinity-mask
There was a problem hiding this comment.
Done. Add the link https://spec.oneapi.io/level-zero/latest/core/PROG.html#affinity-mask instead as latest doc.
2 participants
