[GitHub] [tvm] tingyuzaixiao opened a new issue, #15598: [Bug] the squeeze operator of from_onnx causes the shape error of the result

["tingyuzaixiao (via GitHub)" <gi...@apache.org>]

tingyuzaixiao opened a new issue, #15598:
URL: https://github.com/apache/tvm/issues/15598

     %165 = squeeze(%164, axis=[0]) /* ty=Tensor[(3, 70835, 1), float32] span=Squeeze_4:0:0 */;
     %166 = copy(%164) /* ty=Tensor[(1, 3, 70835, 1), float32] span=Identity_5:0:0 */;
     %167 = squeeze(%164, axis=[0]) /* ty=Tensor[(3, 70835, 1), float32] span=Squeeze_6:0:0 */;
     %168 = copy(%164) /* ty=Tensor[(1, 3, 70835, 1), float32] span=Identity_7:0:0 */;
     %169 = broadcast_to_like(%151, %152) /* ty=Tensor[(1, 8, 70835, 1), float32] span=If_0:0:0 */;
     %170 = broadcast_to_like(%158, %159) /* ty=Tensor[(1, 21, 70835, 1), float32] span=If_1:0:0 */;
     %171 = broadcast_to_like(%165, %166) /* ty=Tensor[(1, 3, 70835, 1), float32] span=If_2:0:0 */;
     %172 = broadcast_to_like(%167, %168) /* ty=Tensor[(1, 3, 70835, 1), float32] span=If_3:0:0 */;
     (%169, %170, %171, %172) /* ty=(Tensor[(1, 8, 70835, 1), float32], Tensor[(1, 21, 70835, 1), float32], Tensor[(1, 3, 70835, 1), float32], Tensor[(1, 3, 70835, 1), float32]) */
   
   Look at the number %172 above，the squeeze operator is optimized into squeeze operator, copy operator and broadcast_to_like operator. When executing squeeze(0),  the result shape is (1, 3, 70835, 1), but (3, 70835, 1) is correct.


-- 
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: commits-unsubscribe@tvm.apache.org.apache.org

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