-
Notifications
You must be signed in to change notification settings - Fork 280
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Tacotron2, yolov3*: benchmark coverage for custom devices. #2230
Conversation
Co-authored-by: Eikan Wang <eikan.wang@intel.com>
Co-authored-by: Eikan Wang <eikan.wang@intel.com>
Co-authored-by: Eikan Wang <eikan.wang@intel.com>
@xuzhao9 , @atalman, This PR is a follow-up of Intel GPU-enabling as described in the RFC(pytorch/pytorch#114723). So far, we have landed Intel GPU runtime and Inductor and limited ATen operations. We are evaluating the dynamo benchmarks and have obtained a reasonable accuracy pass rate for HF. However, for TorchBench, we encountered some functionality issues with the Intel GPU as Intel GPU is a new device, and we need to enable it. This PR intends to enable Intel GPU and the fix the functionality issues. |
The CI failure is pre-existing and is not related to this PR. |
@xuzhao9 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
Works for Roadmap #1293 to increase benchmark coverage.
For these 5 models:tacotron2, yolov3, nvidia_deeprecommender, LearningToPoint and pytorch_CycleGAN_and_pix2pix,
when running on custom devices except for CPU and CUDA(e.g. XPU), it will raise the error as it's hard-coded with CPU/CUDA backends.
In this PR, we accept the device args as a param within the training process and inference process which will cover the model initializing and data transposition for these custom devices.