[GitHub] [incubator-tvm] anijain2305 opened a new pull request #6119: [QNN] Excluding most negative value

[GitBox]

anijain2305 opened a new pull request #6119:
URL: https://github.com/apache/incubator-tvm/pull/6119


   After going through multiple frameworks, I found that it is common to ignore 
-128 for int8 quantization.
   
   TFLite - from Pete Warden's blog and TFLite paper - 
https://petewarden.com/2017/06/22/what-ive-learned-about-neural-network-quantization/
   
   
![image](https://user-images.githubusercontent.com/13822661/88244830-db2c1680-cc49-11ea-9ae6-fd6bf837936c.png)
   
   MKLDNN - https://oneapi-src.github.io/oneDNN/ex_int8_simplenet.html
   
   
![image](https://user-images.githubusercontent.com/13822661/88244853-ee3ee680-cc49-11ea-8efd-5b4ac204ce2d.png)
   
   
   TensorRT - 
https://blog.tensorflow.org/2019/06/high-performance-inference-with-TensorRT.html
   
   
![image](https://user-images.githubusercontent.com/13822661/88244877-00b92000-cc4a-11ea-9909-ea0fd7f6746e.png)
   
   As this is a common practice now, this PR adapts to this change as well.
   
   
   
   


----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org