Hi, I know this is somehow related to another Qualcomm tool (SNPE) but I would love to know if there is a way to somehow load the quantization parameters from AIMET and use it in SNPE?
Or is there a way to modify the quantization parameters in DLC so that I can manually make use of the .encodings file exported from AIMET’s quantization simulation? (I heard from a technical meeting with the dev guys from Qualcomm that there might be a script for this?)
To make things more concrete, I have a quantized model stored in these 2 files: .onnx and .encodings, both exported from AIMET’s quantization simulation. I want to load the model directly to SNPE without SNPE doing the quantization solely based on the ONNX and another input data again.
https://developer.qualcomm.com/docs/snpe/tools.html#tools_snpe-dlc-quantize
The command does support overwriting the quantization parameter in tensorflow. I am using Pytorch and its exported ONNX models.
Greatly appreciate it if you can provide some insight!