I have quantized ONNX model (exported from PyTorch). Is there any way to convert it to quantized TFLite model? It's important to apply quantization on the PyTorch side.
Asked
Active
Viewed 485 times
0
-
did you check - https://siliconlabs.github.io/mltk/mltk/tutorials/onnx_to_tflite.html – Vijay Mariappan Nov 24 '22 at 13:06
-
1Yes, it doesn't work because OpenVino intermediate format uses FP32/FP16 precision. And with this approach, quantization is done at OpenVino -> TFLite conversion – albert828 Nov 24 '22 at 14:22