Frequently Asked Questions (FAQ)

How to give feedback(only for internal users)

1. Access: If AMD internal users cannot access the internal JIRA system, they should first request access permissions following the instructions provided in How to Request Access to JIRA.

2. Reporting: Once access is granted, users should file a JIRA ticket under the project named “quark” in the internal JIRA system. to report an issue or make a request.

Quark for Pytorch

Environment Issues

Issue 1:

Windows CPU mode does not support fp16.

Solution:

Because of torch issue, Windows CPU mode cannot perfectly support fp16.

C++ Compilation Issues

Issue 1:

Stuck in the compilation phase for a long time (over ten minutes), the terminal shows like:

[QUARK-INFO]: Configuration checking start.

[QUARK-INFO]: C++ kernel build directory [cache folder path]/torch_extensions/py39...

Solution:

delete the cache folder [cache folder path]/torch_extensions and run quark again.

Quark for ONNX

Model Issues

Issue 1:

Error of “ValueError:Message onnx.ModelProto exceeds maximum protobuf size of 2GB”

Solution:

This error is caused by the input model size exceeding 2GB. Please set optimize_model=False and use_external_data_format=True.

Quantization Issues

Issue 1:

Error of “onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running Reshape node.”

Solution:

For networks with an ROI head, such as Mask R-CNN or Faster R-CNN, quantization errors may arise if ROIs are not generated in the network. Please use quark.onnx.PowerOfTwoMethod.MinMSE or quark.onnx.CalibrationMethod.Percentile quantization and perform inference with real data.