FNO and TFNO did not the FNO/TFNO blocks #64
Merged
Conversation
This file contains hidden or 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
Contributor
|
Thanks for catching that, we'll include these modifications as we rerun some benchmarks to make sure whatever checkpoints are public match posted numbers. (@kazewong) Setting off the CI for now, but we can merge this as soon as ruff is happy. |
Contributor
|
Great, looks like everything passed off the bat! Thanks again for catching this @AnnihilatorChess . |
Contributor
|
Also, answer to the second point - I would classify that as part of the bug-to-feature pipeline in that we're aware it is the case, but it was not originally intentional and was largely an artifact of co-building the validation and resume pipelines. |
Contributor
Author
|
Is there any update on the benchmarks for the corrected versions? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
Hello,
After some weird behavior of the FNO i discovered that the FNO blocks are not used in both models. The blocks are called but the output is not assigned to x. The same is true for the tfno.
I tested this wrong implementation of the FNO and managed to reproduce the benchmarks for the turbulent_radiative_layer_2D dataset of the paper (one-step VRMSE: ~0.5). After applying this change to actually use the FNO blocks, I tested the FNO with the default settings of the_well for 100 epochs (adjusted LR to 1e-3 and batch_size=8 to fit on my GPU) and achieved VRMSE of ~0.3. This makes me think that the benchmark results in the paper for the turbulent_radiative_layer_2D dataset were also computed with this incorrect implementation.
I can share the results of both runs (wrong implementation and correct implementation) if you want.
On another note that has nothing to do with this pull request. The test set performance is computed with the most recent and not the best model (on the val set). I am not sure whether this is a design choice or if this was overlooked.