Hello everyone,
I have a model that I have been using on my android project.
I am currently trying to upgrade snpe version from 1.66 to 2.10 and I am having problem with quantization.
While using snpe 1.66, I have converted my model on snpe 1.66 using snpe-tensorflow-to-dlc, and I have quantized the model with optimization option cle.
Even though it gave an error message "ERROR Invalid model for HBA quantization algorithm. Ensure there were batchnorms present prior to initial conversion, or reconvert." the quantization worked.
However on snpe 2.10, it gives segmentation fault when I try to quantize the model with the exact same option. The tensorlfow-to-dlc conversion worked fine on snpe 2.10 but only the quantization with cle option does not work. I have been using bc quantized model but I want to use cle quantized model since the snpe documentation states that "In addition, BC should not be run on its own without CLE as it won't fix major issue with weight/activation quantization".
please help
Thank you
Dear developer,
Could you help to withdraw BC/CLE from your SNPE quantization steps?
CLE has been deprecated on latet SNPE verion.
BR.
Wei