Skip to content

Conversation

@natolambert
Copy link
Contributor

cc @patrickvonplaten. Will let you know if I fix this before bed today. Numpy not having a generator makes this tricky.

Some easy fixes are in this, such as get_variance in DDPMTesterClass needs to be changed to _get_variance (what should the common terminology on this be)

Then this change also broke the ve sde test for the full loop, which is very sensitive to changes in randn_like. This is a common problem for these tests.

@HuggingFaceDocBuilderDev
Copy link

HuggingFaceDocBuilderDev commented Jul 20, 2022

The documentation is not available anymore as the PR was closed or merged.

@natolambert
Copy link
Contributor Author

Okay, @patrickvonplaten this should be fine to merge. As long as we are okay with scheduler tests starting in PyTorch, which is a little bit easier to work with given the random sampling of some schedulers and current API.

@natolambert natolambert changed the title work in progress, fixing tests for numpy and make deterministic fixing tests for numpy and make deterministic (ddpm) Jul 20, 2022
@patrickvonplaten patrickvonplaten merged commit fe98574 into main Jul 21, 2022
@patrickvonplaten patrickvonplaten deleted the ddpm_tests branch July 21, 2022 00:25
yoonseokjin pushed a commit to yoonseokjin/diffusers that referenced this pull request Dec 25, 2023
* work in progress, fixing tests for numpy and make deterministic

* make tests pass via pytorch

* make pytorch == numpy test cleaner

* change default tensor format pndm --> pt
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.

4 participants