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.

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

Various SDXL quantization fixes #977

Merged
merged 24 commits into from
Jul 17, 2024
Merged

Various SDXL quantization fixes #977

merged 24 commits into from
Jul 17, 2024

Conversation

nickfraser
Copy link
Collaborator

@nickfraser nickfraser commented Jun 25, 2024

Adds the following fixes / enhancements:

  • Set the device correct when checkpoints are loaded
  • Fix issue setting the bitwidth for the output of linear layers when SDP is quantized
  • Fix replacing the LoRACompatible linear layers
    • Fix or workaround (i.e., achieve this another way) failing tests
  • Move weight to CPU before 'param_only' export
  • Add explicit pandas dependency
  • Add option to scale & export a whitelist of layers of VAE, to enable FP16 execution of VAE

@nickfraser nickfraser marked this pull request as ready for review June 28, 2024 16:15
@nickfraser nickfraser requested review from Giuseppe5 and removed request for Giuseppe5 June 28, 2024 16:22
@Giuseppe5 Giuseppe5 requested review from Giuseppe5 and removed request for Giuseppe5 July 3, 2024 11:36
@@ -287,6 +287,6 @@ def forward(self, x):
model = TestModel()
assert model.conv.stride == (1, 1)

kwargs = {'stride': lambda module: 2 if module.in_channels == 3 else 1}
kwargs = {'stride': lambda module, name: 2 if module.in_channels == 3 else 1}
Copy link
Collaborator Author

Choose a reason for hiding this comment

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

Check if anything in brevitas_examples will be affected by this change.

Copy link
Collaborator

Choose a reason for hiding this comment

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

Lambdas gets inspected to decide what's the correct signature to use, so we have backward compatibility

@nickfraser nickfraser mentioned this pull request Jul 16, 2024
20 tasks
@Giuseppe5 Giuseppe5 merged commit 072a02b into Xilinx:dev Jul 17, 2024
22 checks passed
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.

2 participants