szha commented on issue #19817:
URL:
https://github.com/apache/incubator-mxnet/issues/19817#issuecomment-778661805
Actually I think this bug appears to be non-deterministic. If I run the
script a couple more times I get weird results such as the following, which
happens on both v1.x and on master:
```
python3 snippets/take_bug.py
2.0.0
[13:51:16] ../src/storage/storage.cc:199: Using Pooled (Naive)
StorageManager for CPU
||g_param||_2: 7.51E+16 | Param: convs.1.weight
||g_param||_2: 7.08E+15 | Param: convs.1.bias
||g_param||_2: 8.05E+16 | Param: convs.2.weight
||g_param||_2: 1.39E+17 | Param: convs.2.bias
||g_param||_2: 8.18E+16 | Param: convs.3.weight
||g_param||_2: 8.63E+17 | Param: convs.3.bias
||g_param||_2: 1.38E-04 | Param: dense_out.weight
||g_param||_2: 1.06E-02 | Param: dense_out.bias
-------------------------------------------
------- Grad Info
* ||g||_2: 8.85E+17
* ||g||_1: 3.82E+18
* ||g||_inf: 1.07E+17
||g_param||_2: 2.37E-04 | Param: convs.1.weight
||g_param||_2: 2.29E-05 | Param: convs.1.bias
||g_param||_2: 2.23E-04 | Param: convs.2.weight
||g_param||_2: 1.50E-04 | Param: convs.2.bias
||g_param||_2: 4.26E-04 | Param: convs.3.weight
||g_param||_2: 7.02E-04 | Param: convs.3.bias
||g_param||_2: 1.38E-04 | Param: dense_out.weight
||g_param||_2: 1.06E-02 | Param: dense_out.bias
-------------------------------------------
------- Grad Info
* ||g||_2: 1.06E-02
* ||g||_1: 1.75E-02
* ||g||_inf: 1.06E-02
==== Same outputs?
Y_hat1 - Yhat2 = 0.0000
```
----------------------------------------------------------------
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
