Skip to content

Fix qwen2_5_xnnpack_q8da4w.yaml #13965

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

Merged
merged 1 commit into from
Sep 5, 2025
Merged

Fix qwen2_5_xnnpack_q8da4w.yaml #13965

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
5 changes: 4 additions & 1 deletion examples/models/qwen2_5/config/qwen2_5_xnnpack_q8da4w.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,4 +8,7 @@ model:

backend:
xnnpack:
enabled: True
enabled: True

quantization:
Copy link
Contributor

Choose a reason for hiding this comment

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

For llama, we use

quantization:
  qmode: 8da4w
  group_size: 128
  embedding_quantize: 4,32

Is 8da4w good enough for accuracy?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Yeah we can do that as well. Caveat though that I remember the 0.5B version not reacting well to embedding quantization

Copy link
Contributor

Choose a reason for hiding this comment

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

Ah yeah we don't want embedding to quantize for sure. But I think we need group wise quantization. If we just do qmode: 8da4w, is it using group wise quantization with a default group size?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Yeah, 128 would be the default group size

qmode: 8da4w
Loading