Environment Setup

Q: Why does the error message No module named ‘quantize’ appear while quantize.cpython-35m-x86_64-linux-gnu.so is found in the corresponding directory?

A: The SigmaStar IPU SDK must be used under system Python3.5 environment. Please select the docker environment provided with the SDK for operation.

Q: Why does ImportError: SGS_IPU_SDK/Scripts/calibrator/utils/quantization/ quantize.cpython-35m-x86_64-linux-gnu.so: undefied symbol: PyFPE_jbuf appear?

A: Using Conda environment or non-system Python3.5 may lead to this error. Please select the docker environment provided with the SDK for operation.

Q: Can the docker be set up under Ubuntu or Windows? What is the minimum memory requirement?

A: Yes, we recommend using docker with a minimum of 6G memory, preferably 8G at least. If the memory is just enough, when too many processes are used for calibration, unexpected errors may occur and cause the operation to be interrupted. At present, the operating environment of the developers is based on Ubuntu, and the operation of the docker under Windows is supposed to be okay. However, we are not sure if the docker environment established under Windows is 100% problem-free.

Q: Why the following errors appear when using calibrator_custom?

AttributeError: module 'calibrator_custom' has no attribute 'fixed_simulator' 
AttributeError: module 'calibrator_custom' has no attribute 'offline_simulator'

A: Since fixed-point models and offline models both employ 32-bit runtime library, the module should be operated under 32-bit Python3.5 environment. For this reason, the docker provided has been updated to 32-bit Python3.5 environment, so using python32 should be fine.