Skip to content

Ptq stabilization #72

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

Merged
merged 4 commits into from
May 28, 2020
Merged

Ptq stabilization #72

merged 4 commits into from
May 28, 2020

Conversation

narendasan
Copy link
Collaborator

Description

Stabilizes the C++ calibrator implementation, using the dataloader alone caused non deterministic segfaults, so we now buffer calibration data in the calibrator

Also addresses a bug in the FP16 version of the benchmarking app where a workaround for batch norm in FP16 messed up compilation with the new batch norm implementation.

Now both ptq and benchmark also produce engine files for external testing (if SAVE_ENGINE) is defined.

Type of change

Please delete options that are not relevant and/or add your own.

  • Bug fix (non-breaking change which fixes an issue)
  • New feature (non-breaking change which adds functionality)

Checklist:

  • My code follows the style guidelines of this project
  • I have performed a self-review of my own code
  • I have commented my code, particularly in hard-to-understand areas and hacks
  • I have made corresponding changes to the documentation and have regenerated the documentation (make html in docsrc)
  • New and existing unit tests pass locally with my changes

buffered version to improve stability

Signed-off-by: Naren Dasan <[email protected]>
Signed-off-by: Naren Dasan <[email protected]>
interfere with TRTorch

Signed-off-by: Naren Dasan <[email protected]>
Signed-off-by: Naren Dasan <[email protected]>
benchmarking

Signed-off-by: Naren Dasan <[email protected]>
Signed-off-by: Naren Dasan <[email protected]>
@narendasan narendasan merged commit 40564c3 into master May 28, 2020
@narendasan narendasan deleted the ptq_stabilization branch May 28, 2020 02:04
frank-wei pushed a commit that referenced this pull request Jun 4, 2022
Summary:
Pull Request resolved: https://github.com/pytorch/fx2trt/pull/72

1. support value in pad op
2. permute op improvement since it could not handle x.permute(permutation)
3. improve setitem:
   1) fix issue with slice(-n,None,None)

4. improve getitem:
   1) support x[slice(None,None,None), slice(0,0,None)]
This case is needed for setitem pass optimization. When we need to split a range where the dimension is (20,10)
(slice(None,None,None), slice(-10,None,None)) -->
(slice(None,None,None), slice(0,-10,None)) + (slice(None,None,None), slice(-10,None,None))

Reviewed By: yinghai

Differential Revision: D36223023

fbshipit-source-id: 7d87c74b85b5c6c6efcd46dac209a46af208d3f9
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant