Intermediate time generation #118
Open
+766
−94
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…to intermediate_time_generation
sblackburn86
approved these changes
Jan 22, 2025
| @@ -562,12 +574,17 @@ def generate_samples(self): | |||
| assert ( | |||
| self.hyper_params.diffusion_sampling_parameters is not None | |||
| ), "sampling parameters must be provided to create a generator." | |||
| with torch.no_grad(): | |||
| with ((torch.no_grad())): | |||
| weighted_regularizer_loss = ( | ||
| self.regularizer.compute_weighted_regularizer_loss( | ||
| score_network=self.axl_network, | ||
| augmented_batch=augmented_batch, | ||
| current_epoch=self.current_epoch)) | ||
|
|
||
| t2 = time.time() | ||
| model_regularization_time = t2 - t1 | ||
| logger.info(f" - batch {batch_idx} :: Prediction time = {model_prediction_time:2.1e} s, " |
There was a problem hiding this comment.
is this useful beyond debugging? I'm fine with leaving it in the log, but I'm not sure it adds something.
|
|
||
| def on_train_epoch_start(self) -> None: | ||
| """On train epoch start.""" | ||
| logger.info(f"Starting Training Epoch {self.current_epoch}.") |
There was a problem hiding this comment.
doesn't this overlap with the standard progressbar in the log? We have an option to activate / deactivate it.
|
|
||
| def on_train_epoch_end(self) -> None: | ||
| """On train epoch end.""" | ||
| logger.info(f"Ending Training Epoch {self.current_epoch}.") |
There was a problem hiding this comment.
same as previous comment
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In this PR, I create an experiment to noise a validation data example to an intermediate time and denoise it back again to see when we start having problems. The basic idea is documented in this Notion note: https://www.notion.so/4f446345516e4b499126ad1e97d026cd?v=0618487565864066a0c6c96b4c655d2c&p=17a690083be4809692fef56c24bb97d2&pm=s.
In a nutshell, it lets me create these kinds of images.
Also, I make a few things into
torch.nn.Modulebecause it is a PITA to constantly have to figure out what device we are on. Some of the auxiliary objects we create (such as a regularizer, for example), are created before the whole PL is shipped off to the GPU, so I can't know at creation time if I'm going to CPU or GPU... Better to work with the framework instead of fighting against it, and let PL handle devices.