Introducing structured model outputs #316
Merged
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
…finition While having the argument means we have choice, it does make it more difficult to control the logic
Signed-off-by: Lee, Kin Long Kelvin <[email protected]>
Signed-off-by: Lee, Kin Long Kelvin <[email protected]>
laserkelvin
added
enhancement
|
CI passed on relevant things |
melo-gonzo
approved these changes
Nov 12, 2024
| if isinstance(encoder_outputs, Embeddings): | ||
| embeddings = encoder_outputs | ||
| # for BYO output head cases | ||
| elif isinstance(encoder_outputs, ModelOutput): |
There was a problem hiding this comment.
A comment that may not be super important to consider right at this moment but should be noted - some workflows, such as serving models with OpenKIM's kusp to then run benchmarks, require that node energies are present. Some models (such as mace) can output these directly and they may be worth hanging onto as well.
There was a problem hiding this comment.
I think they're included in ModelOutput - in the wrapper output they're stashed as node_energies. That's what you mean right?
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Outdated suggestions cannot be applied.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.
This PR is semi-related to #104, allowing flexibility for wrappers and full implementations to potentially have a common output format that is also "backwards" compatible. It depends on
ModelOutputfrom #315, so please review and merge that before this PR.AbstractTaskclass (which graph models inherit) called__skip_output_heads__, whichBaseTaskLitModuleswill check this to determine if output heads are needed or not. If we do not use them, we do not initialize theOutputHeadsto save on parameters/ensure we don't have unused parameters.MACEWrapperas an example of how to do this, and we should do so for other wrappers as well (e.g.CHGNet). I've also updated thesphinxdocumentation to detail how this can be used.MACEWrapperto run the forward pass and an end-to-end withForceRegressionTask.