fix(models): processor chunking #629
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
japols
added
bug
japols
commented
Oct 23, 2025
japols
changed the title
anaprietonem
added
ATS Approved
japols
changed the title
Contributor
|
@japols could be worth updating the docs to reflect this change - https://github.com/ecmwf/anemoi-core/blob/main/models/docs/introduction/overview.rst - I guess the section regarding the chunking of the Processors? |
jakob-schloer
requested changes
Nov 14, 2025
Collaborator
jakob-schloer
left a comment
jakob-schloer
left a comment
There was a problem hiding this comment.
Nice work! The processor is much cleaner to read now.
Could you maybe update the documentation to quickly explain what chunks are in the processor and how they relate to layers.
github-project-automation
bot
moved this from To be triaged
to Under Review
in Anemoi-dev
Co-authored-by: Jakob Schloer <[email protected]>
for more information, see https://pre-commit.ci
Co-authored-by: Jakob Schloer <[email protected]>
jakob-schloer
previously approved these changes
Nov 17, 2025
Collaborator
jakob-schloer
left a comment
jakob-schloer
left a comment
There was a problem hiding this comment.
Thanks for updating! Looks good to me now.
anaprietonem
previously approved these changes
Nov 18, 2025
Contributor
anaprietonem
left a comment
anaprietonem
left a comment
There was a problem hiding this comment.
LGTM, I think extending the comment in the transfer learning function to explain the need for applying the sync could be good. I triggered the integration tests - if those pass I think it's good to be merged https://github.com/ecmwf/anemoi-core/actions/workflows/integration-tests-hpc.yml
japols
commented
Nov 19, 2025
anaprietonem
approved these changes
Nov 19, 2025
Draft
OpheliaMiralles
pushed a commit
that referenced
this pull request
Nov 19, 2025
## Description Remove the separate ProcessorChunk class and flatten all layers directly into the BaseProcessor. Chunking is now handled dynamically at runtime by grouping layers into checkpointed segments. ## What problem does this change solve? Previously, the Processor class held a list of ProcessorChunks which held its own ModuleList of layers, meaning that checkpointed layer groupings were tied to the chunking configuration saved in the model checkpoint. When resuming training with a different num_chunks, the restored module structure no longer matched the saved one, causing checkpoint mismatches. Now we only have one flat list of all layers (Blocks) in the Processor Class and chunking is handled dynamically. ## What issue or task does this change relate to? <!-- link to Issue Number --> ## Additional notes ## Tested with all models, i.e. GT, Transformer, GNN, PointWiseMLP ***As a contributor to the Anemoi framework, please ensure that your changes include unit tests, updates to any affected dependencies and documentation, and have been tested in a parallel setting (i.e., with multiple GPUs). As a reviewer, you are also responsible for verifying these aspects and requesting changes if they are not adequately addressed. For guidelines about those please refer to https://anemoi.readthedocs.io/en/latest/*** By opening this pull request, I affirm that all authors agree to the [Contributor License Agreement.](https://github.com/ecmwf/codex/blob/main/Legal/contributor_license_agreement.md) <!-- readthedocs-preview anemoi-training start --> ---- 📚 Documentation preview 📚: https://anemoi-training--629.org.readthedocs.build/en/629/ <!-- readthedocs-preview anemoi-training end --> <!-- readthedocs-preview anemoi-graphs start --> ---- 📚 Documentation preview 📚: https://anemoi-graphs--629.org.readthedocs.build/en/629/ <!-- readthedocs-preview anemoi-graphs end --> <!-- readthedocs-preview anemoi-models start --> ---- 📚 Documentation preview 📚: https://anemoi-models--629.org.readthedocs.build/en/629/ <!-- readthedocs-preview anemoi-models end --> --------- Co-authored-by: Simon Lang <[email protected]> Co-authored-by: gabrieloks <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Ana Prieto Nemesio <[email protected]> Co-authored-by: Jakob Schloer <[email protected]>
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.
Description
Remove the separate ProcessorChunk class and flatten all layers directly into the BaseProcessor.
Chunking is now handled dynamically at runtime by grouping layers into checkpointed segments.
What problem does this change solve?
Previously, the Processor class held a list of ProcessorChunks which held its own ModuleList of layers, meaning that checkpointed layer groupings were tied to the chunking configuration saved in the model checkpoint.
When resuming training with a different num_chunks, the restored module structure no longer matched the saved one, causing checkpoint mismatches. Now we only have one flat list of all layers (Blocks) in the Processor Class and chunking is handled dynamically.
What issue or task does this change relate to?
Additional notes
Tested with all models, i.e. GT, Transformer, GNN, PointWiseMLP
As a contributor to the Anemoi framework, please ensure that your changes include unit tests, updates to any affected dependencies and documentation, and have been tested in a parallel setting (i.e., with multiple GPUs). As a reviewer, you are also responsible for verifying these aspects and requesting changes if they are not adequately addressed. For guidelines about those please refer to https://anemoi.readthedocs.io/en/latest/
By opening this pull request, I affirm that all authors agree to the Contributor License Agreement.
📚 Documentation preview 📚: https://anemoi-training--629.org.readthedocs.build/en/629/
📚 Documentation preview 📚: https://anemoi-graphs--629.org.readthedocs.build/en/629/
📚 Documentation preview 📚: https://anemoi-models--629.org.readthedocs.build/en/629/