Skip to content
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.

Sign up for GitHub

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

Fix AutoAWQQuantizer GptqQuantizer supported ep and device #1571

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix AutoAWQQuantizer GptqQuantizer supported ep and device #1571

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions olive/olive_config.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -228,15 +228,15 @@
},
"AutoAWQQuantizer": {
"module_path": "olive.passes.pytorch.autoawq.AutoAWQQuantizer",
"supported_providers": [ "CPUExecutionProvider" ],
"supported_accelerators": [ "cpu" ],
"supported_providers": [ "CUDAExecutionProvider" ],
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

maybe we should make it * for both providers and ep. the quantization happens on pytorch model and the exported model is compatible with all eps.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

their only requirement is that they require gpus to run. but that is a host machine requirement, not target provider or ep.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this supported_providers mean the output model supports which ep? Same concept for supported_accelerators?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yes. it's used by the auto-opt to filter out the passes based on the intended target ep

Copy link
Contributor Author

@xiaoyu-work xiaoyu-work Jan 24, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. I can see some other passes like QNNConversion is also in this list. But QNN model doesn't not have any ep concept. If this is only used by auto-opt, and auto-opt is targeting onnx model, should those unrelated pass whose output model is not onnx model be removed here?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@shaahji can comment more on it. I think in general, we can just use * for passes that don't deal with onnx models. that way if they are scheduled by the auto-opt, they don't get filtered out.

"supported_accelerators": [ "gpu" ],
"supported_precisions": [ "int4", "int8", "int16", "uint4", "uint8", "uint16" ],
"module_dependencies": [ "autoawq" ]
},
"GptqQuantizer": {
"module_path": "olive.passes.pytorch.gptq.GptqQuantizer",
"supported_providers": [ "CPUExecutionProvider" ],
"supported_accelerators": [ "cpu" ],
"supported_providers": [ "CUDAExecutionProvider" ],
"supported_accelerators": [ "gpu" ],
"supported_precisions": [ "int4", "int8", "int16", "uint4", "uint8", "uint16" ],
"module_dependencies": [ "auto-gptq", "optimum" ]
},
Expand Down
Loading