Parakeet: Support quantization on XNNPACK

[mergennachin]

Old PR: #17175

CI: https://github.com/pytorch/executorch/actions/runs/21691143266/job/62550971847?pr=17216

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17216

• 📄 Preview Python docs built from this PR

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

❌ 1 New Failure, 29 Pending

As of commit 0406ebc with merge base 267a59d ():

NEW FAILURE - The following job has failed:

• trunk / test-mcu-cortex-m-backend / linux-job (gh)
  RuntimeError: Command docker exec -t 099c5b2dcad6e3467bf562382368c4601a2bcf9cd3c6a29285cd2d4ec7f4fbff /exec failed with exit code 1

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

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

[Copilot]

Pull request overview

This PR extends the Parakeet TDT export pipeline to support dynamic quantization for XNNPACK (8da4w), wires it into the ExecuTorch lowering path, documents the workflow, and adds CI coverage.

Changes:

• Updates Parakeet’s quantization config to use intx_choose_qparams_algorithm="hqq_scale_only" for the 8da4w dynamic-activation / 4-bit-weight path.
• Enhances the Parakeet XNNPACK export pipeline by using both dynamic-quant and general XNNPACK partitioners, enabling quant-fusion and const-prop in ExecutorchBackendConfig, and adds documentation and CLI examples for XNNPACK dynamic quantization.
• Extends CI and utility scripts (export_model_artifact.sh, test_model_e2e.sh, pull.yml) to support an xnnpack device mode and a new quantized-8da4w preset for Parakeet, with an end-to-end test job.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
examples/models/parakeet/quantize.py	Adds intx_choose_qparams_algorithm="hqq_scale_only" to the 8da4w Int8DynamicActivationIntxWeightConfig to align Parakeet linear quantization behavior with HQQ-style scale-only parameter selection.
examples/models/parakeet/export_parakeet_tdt.py	Introduces _create_xnnpack_partitioners using both XnnpackDynamicallyQuantizedPartitioner and XnnpackPartitioner, and enables do_quant_fusion_and_const_prop=True in ExecutorchBackendConfig to optimize dynamic quantized XNNPACK exports.
examples/models/parakeet/README.md	Documents a concrete “Dynamic Quantization for XNNPACK” example using --backend xnnpack with 8da4w configs and group size 32.
.github/workflows/pull.yml	Adds a test-parakeet-xnnpack-linux workflow job that exports Parakeet with XNNPACK + quantized-8da4w and runs the e2e test runner.
.ci/scripts/test_model_e2e.sh	Generalizes the script to handle xnnpack as a device, adds the quantized-8da4w option (XNNPACK-only), and maps xnnpack to the *-cpu CMake target so the Parakeet CPU/XNNPACK runner can be built and exercised.
.ci/scripts/export_model_artifact.sh	Adds xnnpack as a supported device, introduces a quantized-8da4w quantization preset restricted to XNNPACK, and routes Parakeet exports on XNNPACK to fp32 (no --dtype override) while preserving bf16 for CUDA/Metal.

---

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

[Copilot]

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated no new comments.

---

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