luyaor opened a new issue #7245:
URL: https://github.com/apache/tvm/issues/7245


   ## Description
   
   When compiling following model with TVM, it will output wrong result.
   
   The model(with ONNX as frontend) with error is as follows, check bug.onnx in 
[bug6.zip](https://github.com/apache/tvm/files/5794682/bug6.zip).
   
   
![image](https://user-images.githubusercontent.com/7541296/104161312-a1381200-542d-11eb-9108-6842e5e0151c.png)
   
   I think it is because TVM relay's wrong implementation for LpPool. In this 
case, p = 1, the result should be 
   L1 norm of [ 0.5186008, -0.2654016] = 
[0.5387418866157532-0.02014109119772911, 
0.6362205147743225-0.9016221165657043], which is 0.7840024, while tvm output 
sum of its element which is 0.25319922.
   
   ## How to reproduce
   
   ### Environment
   
   Python3, with tvm, onnx, onnxruntime
   
   tvm version: 
[`c31e338`](https://github.com/apache/tvm/commit/c31e338d5f98a8e8c97286c5b93b20caee8be602)
 Wed Dec 9 14:52:58 2020 +0900
   
   1. Download [bug6.zip](https://github.com/apache/tvm/files/5794682/bug6.zip)
   2. Run `python check.py`.
   
   
   
   
   
   


----------------------------------------------------------------
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:
[email protected]


Reply via email to