luyaor opened a new issue #7270:
URL: https://github.com/apache/tvm/issues/7270


   ## Description
   
   When compiling following model with TVM, it will output different result 
with onnxruntime. By manually checking, I found the result of TVM may be wrong.
   
   The model(with ONNX as frontend) with error is as follows, check bug.onnx in 
[bug8.zip](https://github.com/apache/tvm/files/5809509/bug8.zip).
   
   
![image](https://user-images.githubusercontent.com/7541296/104472234-b57b3b00-55f6-11eb-862d-6b30e3903656.png)
   
   
   Here is the relay program.
   
   ```
   #[version = "0.0.5"]
   
   def @main(%node1: Tensor[(3, 3, 2, 3, 2, 5), float32]) {
   
    %0 = reshape(%node1, newshape=[-1]);
   
    %1 = abs(%0);
   
    sum(%1, axis=[0], keepdims=True)
   
   }
   ```
   
   I think the shape of correct result should be (1, 540) instead of (1).
   
   
   
   ## How to reproduce
   
   ### Environment
   
   Python3, with tvm, onnx, onnxruntime
   
   tvm version(latest): 
https://github.com/apache/tvm/commit/1f2b40fe371d22aaadc27fc1cc77778b59201f0f
   
   1. Download [bug8.zip](https://github.com/apache/tvm/files/5809509/bug8.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