Let model precision for XPU device align with CUDA #2587
Let model precision for XPU device align with CUDA #2587jianyizh wants to merge 3 commits intopytorch:mainfrom
Conversation
|
Hi @jianyizh! Thank you for your pull request and welcome to our community. Action RequiredIn order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you. ProcessIn order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA. Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks! |
chuanqi129
left a comment
chuanqi129
left a comment
There was a problem hiding this comment.
Please add some sentences in the description about why we need this change and sign the CLA also
torchbenchmark/util/extra_args.py
Outdated
| return True | ||
| if precision == "fp16": | ||
| return model.device == "cuda" and hasattr(model, "enable_fp16") | ||
| return model.device == "cuda" or model.device == "xpu" and hasattr(model, "enable_fp16") |
There was a problem hiding this comment.
| return model.device == "cuda" or model.device == "xpu" and hasattr(model, "enable_fp16") | |
| return model.device in ["cuda", "xpu"] and hasattr(model, "enable_fp16") |
|
Hi @xuzhao9, could you please help to review this PR? |
|
Hi @chuanqi129 , my focus is now moved to Triton and operator level benchmarking at https://github.com/pytorch-labs/tritonbench/tree/main. I can review this PR, but we need to sort out how to proceed other PR reviews with the new Torchbench owner cc @atalman |
jianyizh
had a problem deploying
to
docker-s3-upload
GitHub Actions
Failure
jianyizh
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
|
I don't think the failures in cpu ci are related to this pr. |
Thanks @xuzhao9 for the information and help to review this PR! Hi @atalman could you please help to check the PR check failure and whether we can land this PR? |
|
@atalman Could you please help review this pr |
4 participants
torchbench by default load some models in fp16 if uses gpu. We align such behavior on xpu devices. Also aligned with cuda in nanogpt to use fused adam optimizer