[GitHub] [singa] chrishkchris commented on pull request #763: tensor operation type lang pair - kInt CPU, cuda

[GitBox <gi...@apache.org>]

chrishkchris commented on pull request #763:
URL: https://github.com/apache/singa/pull/763#issuecomment-657429882


   This PR is a problem (bug) fix to issue #744 , and can be merged after I complied and tested it with GPU.
   
   1. test_onnx_backend.py
   
   ```
   root@71ac539cda77:~/dcsysh/singa/test/python# python3 test_onnx_backend.py
   ss............................ssssssssssssssssssssssssssssssss................ssss..ss..ss............ssssssssss..ssssssssssssssss............ssss....................ssssssss........................ssss....ssssssssssssssssssssssssss........ssssssssssssssssssssssssssss..........ssssssssssssss.......s.sssssss....ssssss..................ssss....ssssss..ssss..........ss.s............ssss....ssssssssssssssssssss........ssssssssssss..............ssssssssssssssssssssssss......ss......ssss..ss..............ssssssssss............ssssssssssssssssssssssssss......ssss....ssssssssssssssssss..................................ss........ssssssssssssssss....ssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssss................ssssssssssssssssssssssssssssssss................ssssssssssssssssss....................ssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssss..........ssss......ssss........ssss..............ss..........................................ssssssssssss....................ssssssssssssssssssss....ssssssssssssssssssssssssssssss..................ssss................
   ----------------------------------------------------------------------
   Ran 1114 tests in 1.256s
   
   OK (skipped=603)
   ```
   
   
   2. simple training test (just to see if it introduces other serious bug)
   
   ```
   root@71ac539cda77:~/dcsysh/singa/examples/cnn# python3 train_cnn.py cnn mnist
   Starting Epoch 0:
   Training loss = 582.622375, training accuracy = 0.792789
   Evaluation accuracy = 0.933393, Elapsed Time = 3.197757s
   Starting Epoch 1:
   Training loss = 234.050247, training accuracy = 0.921675
   Evaluation accuracy = 0.956430, Elapsed Time = 3.176632s
   Starting Epoch 2:
   Training loss = 167.452026, training accuracy = 0.944854
   Evaluation accuracy = 0.969351, Elapsed Time = 3.174831s
   Starting Epoch 3:
   Training loss = 141.909851, training accuracy = 0.952591
   Evaluation accuracy = 0.969752, Elapsed Time = 3.174302s
   Starting Epoch 4:
   Training loss = 121.488564, training accuracy = 0.959212
   Evaluation accuracy = 0.970853, Elapsed Time = 3.173884s
   Starting Epoch 5:
   Training loss = 106.590233, training accuracy = 0.965181
   Evaluation accuracy = 0.976362, Elapsed Time = 3.271060s
   Starting Epoch 6:
   Training loss = 97.972565, training accuracy = 0.967399
   Evaluation accuracy = 0.976863, Elapsed Time = 3.177941s
   Starting Epoch 7:
   Training loss = 89.360428, training accuracy = 0.970484
   Evaluation accuracy = 0.973658, Elapsed Time = 3.177838s
   Starting Epoch 8:
   Training loss = 82.499229, training accuracy = 0.972719
   Evaluation accuracy = 0.977865, Elapsed Time = 3.312110s
   Starting Epoch 9:
   Training loss = 79.484062, training accuracy = 0.973252
   Evaluation accuracy = 0.981070, Elapsed Time = 3.180323s
   
   ```


----------------------------------------------------------------
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:
users@infra.apache.org