quic-sanirudh commented on PR #15772: URL: https://github.com/apache/tvm/pull/15772#issuecomment-1731223227
> To create a variable of such type, don't we end up introducing something like `qnn.quantize` op? Yes, I think we would need a quantize/dequantize op anyway, but as you suggested above, we could just add it to regular relax ops, and only the type of input/output tensor would be either regular Tensor or `QuantTensor`. This way, the operators are just the same definition, but legalization would change based on type of tensor instead of a canonicalization step, and we get the advantage of a full quantized operator implementation. I actually did not understand how we might be able to work with just "QDQ representation" with legalization. Would we be doing the work of both `FQ2I` and `Canonicalize` pass together in one shot to get an actual quantized implementation of the operator and schedule that primfunc? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
