Support mx_tensor and enable it's test on Intel GPU#3248
Support mx_tensor and enable it's test on Intel GPU#3248xiaowangintel wants to merge 2 commits intopytorch:mainfrom
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/3248
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ✅ No FailuresAs of commit 64f4151 with merge base 03c2d28 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
| if torch.xpu.is_available(): | ||
| devices.append("xpu") | ||
|
|
||
|
|
There was a problem hiding this comment.
Suggest to use utils function of torchao to get the available device.
There was a problem hiding this comment.
The get_available_devices(https://github.com/pytorch/ao/blob/main/torchao/utils.py#L139) interface can get devices. However, the output includes "cpu" which is not what we need.
| if torch.xpu.is_available(): | ||
| devices.append("xpu") | ||
|
|
||
|
|
There was a problem hiding this comment.
Suggest to use utils function of torchao to get the available device.
6883039 to
64f4151
Compare
|
@jerryzh168 Please help to review, thanks. |
The MXFP8/MXFP4/MXFP6 format training and inference is already enabled on CUDA, but not on Intel GPU. This pr use to support mx function of ao on Intel GPU, and enable the corresponding test. Currently, the linear for mx format is only supported on emulate mode on Intel GPU. We are working the scaled_mm on Intel and will enable more GEMM choice after the scaled_mm on xpu is ready.