mbrookhart commented on pull request #8883: URL: https://github.com/apache/tvm/pull/8883#issuecomment-909352641
Yeah, the way the math works out, we can only really do per-channel quantization on weights and still end up with an integer conv/dense op, if we have per-channel quantization on the data going into one of these ops the lowering breaks. All the binary stuff is to help with fusing ops that show up in the graph after the contraction but before the requantize, i.e., things we want to fuse into the contraction. -- 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]
