Arm backend: TOSAQuantizerV2 fixes

[AdrianLundell]

Break out fixes from #19758 as discussed in #19966

cc @digantdesai@freddan80@per@zingo@oscarandersson8218@mansnils@Sebastian-Larsson@robell@rascani

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20031

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures, 1 Cancelled Job

As of commit 334e6ff with merge base ac3003e ():

NEW FAILURES - The following jobs have failed:

• Test WebGPU Backend / test-webgpu / test-backend-linux (webgpu, models) / linux-job (gh)
  RuntimeError: Command docker exec -t 82c869a9194a5a068b5ceb0a4f347ef832fc77519fa892adc44b2fdf57df5732 /exec failed with exit code 1
• Test WebGPU Backend / test-webgpu / test-backend-linux (webgpu, operators) / linux-job (gh)
  RuntimeError: Command docker exec -t 887ca69b9c42d726ca313a372cdd170557c183e1d06cafceee129677de96476a /exec failed with exit code 1

CANCELLED JOB - The following job was cancelled. Please retry:

• pull / unittest-nxp-neutron / linux-job (gh)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[Copilot]

Pull request overview

This PR applies a set of targeted fixes to the Arm backend’s TOSA/VGF quantization flow (extracted from earlier work), primarily improving correctness of quantization annotations for parameters, shared-qspec clusters, and fixed-qparam inputs, plus small docs/test updates.

Changes:

• Restrict weight/bias annotation to true parameterized ops (conv/linear family) to avoid mis-quantizing constant operands of non-parameter ops.
• Extend/adjust quantization behavior: add support entries for additional ops, add fixed-qparams handling for select trig ops, and improve shared-qspec handling for while_loop with additional_inputs.
• Update Arm getting-started docs and adjust/extend test coverage (including Cortex-M portable int8 shared-qspec op coverage).

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
examples/models/llama/tests/test_export_llama_lib.py	Removes unused torch import and loosens assertions about quantizer internals.
docs/source/backends/arm-vgf/tutorials/vgf-getting-started.md	Clarifies AOT compiler usage for quick tests vs production API.
docs/source/backends/arm-ethos-u/tutorials/ethos-u-getting-started.md	Same tutorial clarification as VGF doc.
backends/cortex_m/test/misc/test_portable_int8.py	Adds shared-qspec coverage for additional indexing ops (index_put_, index_copy, etc.).
backends/arm/scripts/docgen/docgen.py	Refines docstring-to-bullets conversion to only bullet first-level arg lines.
backends/arm/quantizer/quantizer_support.py	Updates supported pattern list (adds RNN/LSTM/GRU, trig ops, einsum; removes hardsigmoid from fused activations).
backends/arm/quantizer/quantization_config.py	Adds fixed-qparams input qspec selection for specific trig ops via FixedQParamsQuantizationSpec.
backends/arm/quantizer/quantization_annotator.py	Tightens typing for op sets and exposes fixed-input-qspec mapping used by config.
backends/arm/quantizer/arm_quantizer_utils.py	Fixes parameter handling by identifying weights/biases by actual use (arg position + allowed targets) and adds while_loop shared-qspec handling.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rascani]

The failing size tests likely just need a branch update. WebGPU are pre-existing.

[AdrianLundell]

Break out fixes from #19758 as discussed in #19966

cc @digantdesai@freddan80@per@zingo@oscarandersson8218@mansnils@Sebastian-Larsson@robell@rascani

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20031

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures, 1 Cancelled Job

As of commit 334e6ff with merge base ac3003e ():

NEW FAILURES - The following jobs have failed:

• Test WebGPU Backend / test-webgpu / test-backend-linux (webgpu, models) / linux-job (gh)
  RuntimeError: Command docker exec -t 82c869a9194a5a068b5ceb0a4f347ef832fc77519fa892adc44b2fdf57df5732 /exec failed with exit code 1
• Test WebGPU Backend / test-webgpu / test-backend-linux (webgpu, operators) / linux-job (gh)
  RuntimeError: Command docker exec -t 887ca69b9c42d726ca313a372cdd170557c183e1d06cafceee129677de96476a /exec failed with exit code 1

CANCELLED JOB - The following job was cancelled. Please retry:

• pull / unittest-nxp-neutron / linux-job (gh)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[Copilot]

Pull request overview

This PR applies a set of targeted fixes to the Arm backend’s TOSA/VGF quantization flow (extracted from earlier work), primarily improving correctness of quantization annotations for parameters, shared-qspec clusters, and fixed-qparam inputs, plus small docs/test updates.

Changes:

• Restrict weight/bias annotation to true parameterized ops (conv/linear family) to avoid mis-quantizing constant operands of non-parameter ops.
• Extend/adjust quantization behavior: add support entries for additional ops, add fixed-qparams handling for select trig ops, and improve shared-qspec handling for while_loop with additional_inputs.
• Update Arm getting-started docs and adjust/extend test coverage (including Cortex-M portable int8 shared-qspec op coverage).

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
examples/models/llama/tests/test_export_llama_lib.py	Removes unused torch import and loosens assertions about quantizer internals.
docs/source/backends/arm-vgf/tutorials/vgf-getting-started.md	Clarifies AOT compiler usage for quick tests vs production API.
docs/source/backends/arm-ethos-u/tutorials/ethos-u-getting-started.md	Same tutorial clarification as VGF doc.
backends/cortex_m/test/misc/test_portable_int8.py	Adds shared-qspec coverage for additional indexing ops (index_put_, index_copy, etc.).
backends/arm/scripts/docgen/docgen.py	Refines docstring-to-bullets conversion to only bullet first-level arg lines.
backends/arm/quantizer/quantizer_support.py	Updates supported pattern list (adds RNN/LSTM/GRU, trig ops, einsum; removes hardsigmoid from fused activations).
backends/arm/quantizer/quantization_config.py	Adds fixed-qparams input qspec selection for specific trig ops via FixedQParamsQuantizationSpec.
backends/arm/quantizer/quantization_annotator.py	Tightens typing for op sets and exposes fixed-input-qspec mapping used by config.
backends/arm/quantizer/arm_quantizer_utils.py	Fixes parameter handling by identifying weights/biases by actual use (arg position + allowed targets) and adds while_loop shared-qspec handling.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rascani]

The failing size tests likely just need a branch update. WebGPU are pre-existing.

[AdrianLundell]

Break out fixes from #19758 as discussed in #19966

cc @digantdesai@freddan80@per@zingo@oscarandersson8218@mansnils@Sebastian-Larsson@robell@rascani

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20031

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures, 1 Cancelled Job

As of commit 334e6ff with merge base ac3003e ():

NEW FAILURES - The following jobs have failed:

• Test WebGPU Backend / test-webgpu / test-backend-linux (webgpu, models) / linux-job (gh)
  RuntimeError: Command docker exec -t 82c869a9194a5a068b5ceb0a4f347ef832fc77519fa892adc44b2fdf57df5732 /exec failed with exit code 1
• Test WebGPU Backend / test-webgpu / test-backend-linux (webgpu, operators) / linux-job (gh)
  RuntimeError: Command docker exec -t 887ca69b9c42d726ca313a372cdd170557c183e1d06cafceee129677de96476a /exec failed with exit code 1

CANCELLED JOB - The following job was cancelled. Please retry:

• pull / unittest-nxp-neutron / linux-job (gh)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[Copilot]

Pull request overview

This PR applies a set of targeted fixes to the Arm backend’s TOSA/VGF quantization flow (extracted from earlier work), primarily improving correctness of quantization annotations for parameters, shared-qspec clusters, and fixed-qparam inputs, plus small docs/test updates.

Changes:

• Restrict weight/bias annotation to true parameterized ops (conv/linear family) to avoid mis-quantizing constant operands of non-parameter ops.
• Extend/adjust quantization behavior: add support entries for additional ops, add fixed-qparams handling for select trig ops, and improve shared-qspec handling for while_loop with additional_inputs.
• Update Arm getting-started docs and adjust/extend test coverage (including Cortex-M portable int8 shared-qspec op coverage).

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
examples/models/llama/tests/test_export_llama_lib.py	Removes unused torch import and loosens assertions about quantizer internals.
docs/source/backends/arm-vgf/tutorials/vgf-getting-started.md	Clarifies AOT compiler usage for quick tests vs production API.
docs/source/backends/arm-ethos-u/tutorials/ethos-u-getting-started.md	Same tutorial clarification as VGF doc.
backends/cortex_m/test/misc/test_portable_int8.py	Adds shared-qspec coverage for additional indexing ops (index_put_, index_copy, etc.).
backends/arm/scripts/docgen/docgen.py	Refines docstring-to-bullets conversion to only bullet first-level arg lines.
backends/arm/quantizer/quantizer_support.py	Updates supported pattern list (adds RNN/LSTM/GRU, trig ops, einsum; removes hardsigmoid from fused activations).
backends/arm/quantizer/quantization_config.py	Adds fixed-qparams input qspec selection for specific trig ops via FixedQParamsQuantizationSpec.
backends/arm/quantizer/quantization_annotator.py	Tightens typing for op sets and exposes fixed-input-qspec mapping used by config.
backends/arm/quantizer/arm_quantizer_utils.py	Fixes parameter handling by identifying weights/biases by actual use (arg position + allowed targets) and adds while_loop shared-qspec handling.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rascani]

The failing size tests likely just need a branch update. WebGPU are pre-existing.