Skip to content

Sample fix#3

Open
jesperdlau wants to merge 2 commits intoprescient-design:mainfrom
jesperdlau:sample-fix
Open

Sample fix#3
jesperdlau wants to merge 2 commits intoprescient-design:mainfrom
jesperdlau:sample-fix

Conversation

@jesperdlau
Copy link

I encountered some bugs while sampling. Maybe it's just me, but when instantiating the model in
src/walkjump/cmdline/_sample.py, The instantiate_model_for_sample_mode function doesn't seem to accept the DictConfig input. This is solved by either giving the specific inputs like
def instantiate_model_for_sample_mode( model_type, checkpoint_path, denoise_path) -> NoiseEnergyModel | DenoiseModel:

Or by instantiating the model by calling the function directly, like what is done with instantiate_redesign_mask and instantiate_seeds.

I also added some examples and comments.

@justinlubin
Copy link

Hey, I just wanted to say that this was a really helpful PR for me! I could get the model up and running with the changes you suggested. Thanks!

@justinlubin
Copy link

justinlubin commented Apr 16, 2024

I also figured out that you need to specify the following option or something similar (e.g. in src/walkjump/hydra_config/sample.yaml) to enable sampling:

redesign_regions: ["L1", "L2", "H1", "H2"]

Otherwise, every index gets masked, and sampling never changes the seed values!

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