Skip to content

Fix quark quantization PHX/HPT NPU options#379

Open
Rohithmatham12 wants to merge 1 commit into
amd:mainfrom
Rohithmatham12:fix-quark-phx-provider-options
Open

Fix quark quantization PHX/HPT NPU options#379
Rohithmatham12 wants to merge 1 commit into
amd:mainfrom
Rohithmatham12:fix-quark-phx-provider-options

Conversation

@Rohithmatham12

Copy link
Copy Markdown

Summary\n- Align quark_quantization benchmark VitisAI EP cache options with the working int8 ResNet predict path by using cache_dir/cache_key.\n- Set xlnx_enable_py3_round=0 for PHX/HPT alongside the required target and xclbin options.\n- Update the README NPU provider-options snippet so users copy the same option style.\n\nWhy\nIssue #376 reports that CNN-examples/quark_quantization/quark_quantize.py falls back to CPU on PHX/HPT while the same quantized model runs on the NPU through CNN-examples/getting_started_resnet/int8/predict.py. The benchmark path was using different provider option keys and missing the PHX/HPT rounding option used by the working path.\n\nTesting\n- python3 -m py_compile CNN-examples/quark_quantization/quark_quantize.py CNN-examples/quark_quantization/utils.py\n- git diff --check\n\nNot run locally: PHX/HPT NPU benchmark validation, because this machine does not have Ryzen AI hardware/XRT.\n\nFixes #376

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant