[QNN] How to decompose qnn.requantize into relay-only operators

Questions
#1

In the Deploy a Framework-prequantized Model with TVM - Part 3 (TFLite) example, I can get qnn.requantize in the graph. However, I don’t know how to QNN passes to decompose the qnn.requantize. Is there an example for that?

#2

@anijain2305 could you help on this question?

#3

If you are in python, you can just try

mod = relay.qnn.transform.CanonicalizeOps()(mod)

1 Like
Build with opt_level=0 getting error qnn.requantize not registed TOpPattern not registered
#4

Thank you @comaniac @anijain2305 it works.