cee1 commented on PR #11155: URL: https://github.com/apache/tvm/pull/11155#issuecomment-1118130717
> Please add a test for this - what's the impact of accuracy if you compare this with Tensorflow Lite ? > > Are we trading accuracy in one framework for another ? > > @Mousius , @manupa-arm , please could you take a look ? > > Ramana If recall correctly, current TVM's NEON path is bit-exact with output of tflite? But is inconsistent with the fake quantization Op within QAT, which, IMHO, can be seem as the reference implementation of requantization / quantization (where `q_multiply_shift` is involved in) To solve the concern, this patch is submitted. -- 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]
