-
Couldn't load subscription status.
- Fork 702
Fix torchao quantization for deployment targets under iOS18 #13896
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
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/13896
Note: Links to docs will display an error until the docs builds have been completed. ❌ 3 New FailuresAs of commit fc4e8fe with merge base b02db12 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This PR needs a
|
|
Fixes #13122 |
|
FYI to @abhinaykukkadapu. Per-channel 8-bit TorchAO quantization is now supported on iOS16+ with this PR. We can follow up with recipe update. |
|
Passing reminder to use |
12430d4 to
fc4e8fe
Compare
|
@cccclai @digantdesai @abhinaykukkadapu @jackzhxng I need a stamp from someone in PyTorch to merge. The Apple engineer has already approved. |
…tion to IO16 Summary: Following up with recipe update as mentioned here: pytorch#13896 (comment) Differential Revision: D82048153
…tion to IO16 (pytorch#14127) Summary: Following up with recipe update as mentioned here: pytorch#13896 (comment) Reviewed By: metascroy Differential Revision: D82048153
…tion to IO16 (pytorch#14127) Summary: Pull Request resolved: pytorch#14127 Following up with recipe update as mentioned here: pytorch#13896 (comment) Reviewed By: metascroy Differential Revision: D82048153
…tion to IO16 (pytorch#14127) Summary: Following up with recipe update as mentioned here: pytorch#13896 (comment) Reviewed By: metascroy Differential Revision: D82048153
…tion to IO16 (pytorch#14127) Summary: Following up with recipe update as mentioned here: pytorch#13896 (comment) Reviewed By: metascroy Differential Revision: D82048153
…tion to IO16 (pytorch#14127) Summary: Following up with recipe update as mentioned here: pytorch#13896 (comment) Reviewed By: metascroy Differential Revision: D82048153
_construct_constexpr_dequant_op should use axis = None (default value), rather than -1. This will infer the axis from the shape of scales.
On iOS18, this was not an issue because the axis is not used. For iOS16, the axis is used and the value -1 usually caused an issue during export.