https://github.com/qqaatw/pytorch-mps-ops-coverage

PyTorch MPS backend Operators Coverage
https://github.com/qqaatw/pytorch-mps-ops-coverage

Last synced: 23 days ago
JSON representation

PyTorch MPS backend Operators Coverage

• Host: GitHub
• URL: https://github.com/qqaatw/pytorch-mps-ops-coverage
• Owner: qqaatw
• License: apache-2.0
• Created: 2023-04-03T10:30:33.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2024-10-18T01:50:06.000Z (29 days ago)
• Last Synced: 2024-10-20T14:10:51.853Z (27 days ago)
• Language: HTML
• Size: 82 KB
• Stars: 7
• Watchers: 2
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# PyTorch MPS Operators Coverage

 

Visit https://qqaatw.github.io/pytorch-mps-ops-coverage/ to check the table.

The table is updated automatically every day.

## Note

The table only contains ops under MPS dispatch key in `native_functions.yaml` for simplicity. Manually registered ops must be updated in [config.yml](config.yml).

For other dispatch keys and ops outside PyTorch:

1. `CompositeExplicitAutograd`: Most ops under this dispatch key have been supported by MPS.

2. `CompositeImplicitAutograd`: Most ops under this dispatch key have been supported by MPS.

3. `torchvision`: Except for `deform_conv2d`, other ops such as `nms` have been supported by MPS.

## Contributions

Contributions are welcomed for completing the details of MPS ops coverage in [config.yml](config.yml).

Currently, there are three entries for each op that can be filled:

1. `starting_macOS_version`: The macOS version which starts supporting the op.

2. `note`: A note which describes any worth mentioning details about the op.

3. `included_in_latest`: For manually registered ops, use this to indicate whether the ops are included in the latest PyTorch release.