chrishkchris commented on issue #588: [WIP] SINGA-505 SoftMaxBackward using CUDNN URL: https://github.com/apache/singa/pull/588#issuecomment-580657211 I have tried the CUDNN SoftMax backward by modifying the mnist example: ``` ubuntu@ip-172-31-24-48:~/singa/examples/autograd$ python3 mnist_cnn.py Starting Epoch 0: Training loss = 580.753418, training accuracy = 0.797208 Evaluation accuracy = 0.939002, Elapsed Time = 2.541010s Starting Epoch 1: Training loss = 229.544083, training accuracy = 0.924260 Evaluation accuracy = 0.958133, Elapsed Time = 2.517823s Starting Epoch 2: Training loss = 165.276779, training accuracy = 0.945454 Evaluation accuracy = 0.974559, Elapsed Time = 2.523124s Starting Epoch 3: Training loss = 134.344086, training accuracy = 0.955593 Evaluation accuracy = 0.974159, Elapsed Time = 2.523674s Starting Epoch 4: Training loss = 115.716629, training accuracy = 0.961730 Evaluation accuracy = 0.979367, Elapsed Time = 2.522673s Starting Epoch 5: Training loss = 104.472374, training accuracy = 0.964831 Evaluation accuracy = 0.978265, Elapsed Time = 2.535722s Starting Epoch 6: Training loss = 95.322929, training accuracy = 0.968283 Evaluation accuracy = 0.984575, Elapsed Time = 2.525273s Starting Epoch 7: Training loss = 88.591621, training accuracy = 0.970184 Evaluation accuracy = 0.981571, Elapsed Time = 2.526018s Starting Epoch 8: Training loss = 83.001053, training accuracy = 0.971685 Evaluation accuracy = 0.981571, Elapsed Time = 2.525087s Starting Epoch 9: Training loss = 76.832161, training accuracy = 0.974653 Evaluation accuracy = 0.979267, Elapsed Time = 2.527286s ```
---------------------------------------------------------------- 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: us...@infra.apache.org With regards, Apache Git Services
