Da custom layer #1429
Merged
Da custom layer #1429
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
|
Looks very nice, but I do have a question about the user interface. The switches for the fallback are added to the |
Contributor
Author
|
**kwargs are passed through from convert_from_keras_model |
Contributor
|
Yes, the point is more that they are not explained in the doc for that function. So the user has no pointers on how to configure this, except by reading the code. I guess the specific request would be to add this to the function doc string for |
JanFSchulte
approved these changes
Feb 4, 2026
6 tasks
Abubakar-rashid
added a commit
to Abubakar-rashid/hls4ml
that referenced
this pull request
Feb 6, 2026
Added allow_da_fallback and allow_v2_fallback args from PR fastmachinelearning#1429
JanFSchulte
pushed a commit
that referenced
this pull request
Feb 9, 2026
* Fix Keras v3 model conversion in numerical profiling * Address review feedback: remove unnecessary else clause and add test to LONGLIST - Removed else clause in convert_from_config() as it's no longer needed - Added test_keras_v3_profiling to LONGLIST in generate_ci_yaml.py for proper CI environment * Fix line length issue in generate_ci_yaml.py Break KERAS3_LIST into multiple lines to comply with max line length of 125 characters * Einsum and EinsumDense for oneAPI * Add tests * Add tests * Remove hgq2 from testing-keras3 dependencies and test_hgq2_mha from KERAS3_LIST As requested by reviewers: - Removed hgq2>=0.0.1 from testing-keras3 optional dependencies - Removed test_hgq2_mha from KERAS3_LIST * [pre-commit.ci] pre-commit autoupdate (#1425) updates: - [github.com/astral-sh/ruff-pre-commit: v0.14.13 → v0.14.14](astral-sh/ruff-pre-commit@v0.14.13...v0.14.14) Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> * Add LHC trigger use case context to README (#1418) * Add LHC trigger use case context to README * Clarify hls4ml application domains in README --------- Co-authored-by: Siddardha Desu <siddardhadesu@Siddardhas-MacBook-Air.local> * [pre-commit.ci] pre-commit autoupdate (#1430) updates: - [github.com/tox-dev/pyproject-fmt: v2.11.1 → v2.12.1](tox-dev/pyproject-fmt@v2.11.1...v2.12.1) Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> * Fix Keras v3 profiling tests: build models and correct bar count expectations - Call model.build() or pass sample data to initialize Keras v3 models - Fix bar count assertions to match actual behavior (1 bar per layer, not per weight) - Ensures models have defined inputs before profiling * Fix qkeras import to avoid namespace conflict with hgq2 - Change 'import qkeras' to 'from qkeras import QActivation' in profiling.py - This protects against hgq2's qkeras compatibility layer causing conflicts - Add back hgq2>=0.0.1 to testing-keras3 dependencies - Add back test_hgq2_mha to KERAS3_LIST As requested by reviewers to resolve qkeras/hgq2 namespace issues. * Fix activation and batch norm tests - Call model with data instead of just build() for activation test - Fix batch norm expected bar count to 3 * Add new arguments to convert_from_keras_model call Added allow_da_fallback and allow_v2_fallback args from PR #1429 * Fix Keras v3 profiling tests - 4/5 passing One test (hls_model comparison) is skipped because get_unoptimized_hlsmodel() tries to recreate the model from saved config, which fails for Keras v3 due to config format changes. This is a library issue in keras_v2_to_hls parser that needs a separate fix. --------- Co-authored-by: Jovan Mitrevski <jmitrevs@fnal.gov> Co-authored-by: laurilaatu <l.laatu@imperial.ac.uk> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: siddardhadesu <153359094+siddardhadesu@users.noreply.github.com> Co-authored-by: Siddardha Desu <siddardhadesu@Siddardhas-MacBook-Air.local>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
Description
Allows the use of any subclass of
keras.layers.Layerobject that is supported in da4ml to be used in the hls4ml project, such as LUT-based layers (such ashgq.layers.DenseT/ConvT12Dorkeras.layers.Conv3D). Due to parser limitation,keras.operation.Operationstill cannot be supported (yet), so arbitrary tensor slicing or indexing operations will not work for now.Now, the keras v3 parser will try da parser fallback before trying v2 parser. If it falls by not finding the corresponding layer or da4ml is not present, it will proceed with the current behavior.
Type of change
Tests
Still no keras v3 test infra there.
Checklist