Qualcomm AI Engine Direct - fix LPBQ implementation

[haowhsu-quic]

Summary

• fix LPBQ and make test case more general

Test plan

python backends/qualcomm/tests/test_qnn_delegate.py TestQNNQuantizedOperator.test_qnn_backend_conv2d_block -b build-android -s $DEVICE -m SM8750

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/12663

• 📄 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

As of commit e8ad73c with merge base 4d7f9ca ():

NEW FAILURES - The following jobs have failed:

• pull / android / run-emulator (gh)
  The process '/usr/bin/sh' failed with exit code 255
• pull / test-eval_llama-mmlu-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 7ef489acdfb86e238b3eddd619e248be90969d2c74a6411f83dbf4792939d728 /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.

[haowhsu-quic]

Hi @cccclai, I made a mistake when implementing LPBQ. I've tested current version against Llama3.2 1B and can get better results now. Will update the related change on llama.py in another PR.

[cccclai]

Hi @cccclai, I made a mistake when implementing LPBQ. I've tested current version against Llama3.2 1B and can get better results now. Will update the related change on llama.py in another PR.

Thanks! Mind sharing a bit more details about the mistake? Trying to follow the code but not super clear

[haowhsu-quic]

Yes, currently HTP uses 4bits to store quantized scales. I didn't clip the numeric to the right range (before: 0-255 / after: 1-16). I also double check with AIMET's implementation and the mse between per-channel / per-block.

[cccclai]

Yes, currently HTP uses 4bits to store quantized scales. I didn't clip the numeric to the right range (before: 0-255 / after: 1-16). I also double check with AIMET's implementation and the mse between per-channel / per-block.

Ah that makes sense..

[cccclai]

Thanks for the fix

[facebook-github-bot]

@cccclai has imported this pull request. If you are a Meta employee, you can view this in D78635428.