Arm backend: Add experimental support for new TOSAQuantizer#18100
Open
AdrianLundell wants to merge 4 commits intopytorch:mainfrom
Open
Arm backend: Add experimental support for new TOSAQuantizer#18100AdrianLundell wants to merge 4 commits intopytorch:mainfrom
AdrianLundell wants to merge 4 commits intopytorch:mainfrom
Conversation
Allows initializing TOSA/EthosU/Vgf quantizers with use_composable_quantizer=True to use a new implementation of the quantizer following the Cortex-M. See pytorch#17701 for more details. - Creates a new temporary TOSAQuantizer API layer for switching between the two versions - Adds a TOSAQuantizationConfig encapturing TOSA-specific qspec requirements for certain ops. - Adds quantizer_support.py for defining what operators are supported by the quantizer. - Align mark_node_as_annotated in cortex-m backend to TOSAQuantizer behaviour. - Update quantizer reporter to handle TOSA qspecs as they are dynamically created. Signed-off-by: Adrian Lundell <adrian.lundell@arm.com> Change-Id: Icbca66ff86e6f78ffa1c8dcec55e17c25f97d8ca
AdrianLundell
added
partner: arm
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18100
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New FailuresAs of commit 5c02aa7 with merge base 096f10c ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
Signed-off-by: Adrian Lundell <adrian.lundell@arm.com> Change-Id: Id81e0c39d13a94a749206441fce60664c80a0af8
Collaborator
|
Hi @SS-JIA / @digantdesai this adds a file, do you want/need to check this? |
Collaborator
Author
|
Fails unrelated |
rascani
approved these changes
Mar 12, 2026
|
|
||
| # Lazily import heavy quantizer classes to avoid circular imports with | ||
| # Cortex-M quantization configs. | ||
| _LAZY_EXPORTS = { |
| (torch.ops.aten.sqrt.default,), | ||
| (torch.ops.aten.silu.default,), | ||
| (torch.ops.aten.silu_.default,), | ||
| (torch.ops.aten.logit.default,), |
Contributor
There was a problem hiding this comment.
torch.ops.aten.logit.default listed twice (also at line 126).
| if qspec.is_dynamic != key.is_dynamic: | ||
| continue | ||
| return val | ||
| return "UNREGISTRED_QSPEC" |
| targeted_nodes_description = str(self.node_finder) | ||
| quantization_config_path = SUPPORTED_QCONFIGS.get( | ||
| self.quantization_config, "CUSTOM_QCONFIG" | ||
| self.quantization_config, "UNREGISTRED_QCONFIG" |
Contributor
There was a problem hiding this comment.
Typo: "UNREGISTERED_QCONFIG"
| continue | ||
| if qspec.is_dynamic != key.is_dynamic: | ||
| continue | ||
| return val |
Contributor
There was a problem hiding this comment.
Should we add a check for qspec.ch_axis?
| "//executorch/backends/arm:constants", | ||
| "//executorch/backends/arm:ethosu", | ||
| "//executorch/backends/arm:vgf", | ||
| "//executorch/backends/cortex_m/quantizer:quantizer", |
Contributor
There was a problem hiding this comment.
Pulling this in exposed that the cortex_m/quantizer:quantizer is broken on main. Here's summary of what needs to change:
backends/cortex_m/quantizer/TARGETS — add missing srcs and deps:
python_library(
name = "quantizer",
srcs = [
"__init__.py",
"node_finders.py", # MISSING
"operator_configs.py", # NEEDS REMOVING
"pattern_checkers.py", # MISSING
"pattern_matcher.py", # MISSING
"quantization_configs.py",
"quantizer.py",
"quantizer_reporter.py", # MISSING
"quantizer_support.py", # MISSING
],
deps = [
"//caffe2:torch",
"//executorch/backends/arm:common", # MISSING
"//executorch/backends/arm:constants", # MISSING
"//executorch/backends/arm/quantizer:quantization_annotator", # MISSING
"//executorch/backends/arm/quantizer:quantization_config",
"//pytorch/ao:torchao",
"fbsource//third-party/pypi/tabulate:tabulate", # MISSING
],
)
I'll put together a separate PR, but feel free to incorporate into this one to unblock.
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
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.
Allows initializing TOSA/EthosU/Vgf quantizers with use_composable_quantizer=True to use a new implementation of the quantizer following the Cortex-M. See
#17701 for more details.
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell