chrishkchris edited a comment on issue #651: Add new example APIs URL: https://github.com/apache/singa/pull/651#issuecomment-609758264 I added two args in train_mpi,py and train_multiprocess.py to support all the distributed training options, but train.py don't need the argument because it doesn't have DistOpt Below are four examples for the use of arguments: ``` root@71ac539cda77:~/dcsysh/singa/examples/autograd# mpiexec -np 2 python3 train_mpi.py cnn mnist --op fp16 --lr 0.01 Starting Epoch 0: Training loss = 625.136963, training accuracy = 0.776893 Evaluation accuracy = 0.939704, Elapsed Time = 1.619492s Starting Epoch 1: Training loss = 235.280151, training accuracy = 0.920339 Evaluation accuracy = 0.947115, Elapsed Time = 1.517764s Starting Epoch 2: Training loss = 171.311310, training accuracy = 0.942508 Evaluation accuracy = 0.970252, Elapsed Time = 1.557606s Starting Epoch 3: Training loss = 139.594086, training accuracy = 0.953342 Evaluation accuracy = 0.972356, Elapsed Time = 1.541372s Starting Epoch 4: Training loss = 120.380058, training accuracy = 0.959852 Evaluation accuracy = 0.971655, Elapsed Time = 1.560016s Starting Epoch 5: Training loss = 104.767105, training accuracy = 0.965345 Evaluation accuracy = 0.979467, Elapsed Time = 1.525147s Starting Epoch 6: Training loss = 98.995010, training accuracy = 0.966730 Evaluation accuracy = 0.976963, Elapsed Time = 1.528173s Starting Epoch 7: Training loss = 88.012024, training accuracy = 0.970202 Evaluation accuracy = 0.977865, Elapsed Time = 1.515204s Starting Epoch 8: Training loss = 86.540741, training accuracy = 0.970519 Evaluation accuracy = 0.975361, Elapsed Time = 1.604774s Starting Epoch 9: Training loss = 80.653885, training accuracy = 0.973024 Evaluation accuracy = 0.982071, Elapsed Time = 1.633480s root@71ac539cda77:~/dcsysh/singa/examples/autograd# mpiexec -np 2 python3 train_mpi.py cnn mnist --op partialUpdate --lr 0.01 Starting Epoch 0: Training loss = 623.692139, training accuracy = 0.777694 Evaluation accuracy = 0.939804, Elapsed Time = 1.678683s Starting Epoch 1: Training loss = 235.369232, training accuracy = 0.920690 Evaluation accuracy = 0.948017, Elapsed Time = 1.567066s Starting Epoch 2: Training loss = 171.335693, training accuracy = 0.942157 Evaluation accuracy = 0.971855, Elapsed Time = 1.595140s Starting Epoch 3: Training loss = 139.396088, training accuracy = 0.953592 Evaluation accuracy = 0.971755, Elapsed Time = 1.661490s Starting Epoch 4: Training loss = 120.173752, training accuracy = 0.960019 Evaluation accuracy = 0.970753, Elapsed Time = 1.571710s Starting Epoch 5: Training loss = 105.472672, training accuracy = 0.965061 Evaluation accuracy = 0.978065, Elapsed Time = 1.570710s Starting Epoch 6: Training loss = 99.055389, training accuracy = 0.966930 Evaluation accuracy = 0.977163, Elapsed Time = 1.561047s Starting Epoch 7: Training loss = 88.166275, training accuracy = 0.970002 Evaluation accuracy = 0.976663, Elapsed Time = 1.665272s Starting Epoch 8: Training loss = 85.920563, training accuracy = 0.970469 Evaluation accuracy = 0.976162, Elapsed Time = 1.579652s Starting Epoch 9: Training loss = 79.568375, training accuracy = 0.973040 Evaluation accuracy = 0.982472, Elapsed Time = 1.605414s root@71ac539cda77:~/dcsysh/singa/examples/autograd# mpiexec -np 2 python3 train_mpi.py cnn mnist --op sparseTopK --spars 0.05 --lr 0.01 Starting Epoch 0: Training loss = 1011.997559, training accuracy = 0.637720 Evaluation accuracy = 0.890124, Elapsed Time = 2.135889s Starting Epoch 1: Training loss = 359.958496, training accuracy = 0.876920 Evaluation accuracy = 0.931390, Elapsed Time = 1.748328s Starting Epoch 2: Training loss = 259.234558, training accuracy = 0.912660 Evaluation accuracy = 0.953826, Elapsed Time = 1.694958s Starting Epoch 3: Training loss = 206.229309, training accuracy = 0.930939 Evaluation accuracy = 0.954327, Elapsed Time = 1.627520s Starting Epoch 4: Training loss = 176.526505, training accuracy = 0.940138 Evaluation accuracy = 0.958634, Elapsed Time = 1.618325s Starting Epoch 5: Training loss = 158.930298, training accuracy = 0.947349 Evaluation accuracy = 0.962941, Elapsed Time = 1.619869s Starting Epoch 6: Training loss = 145.359879, training accuracy = 0.951289 Evaluation accuracy = 0.966446, Elapsed Time = 1.621834s Starting Epoch 7: Training loss = 129.951813, training accuracy = 0.955746 Evaluation accuracy = 0.972055, Elapsed Time = 1.712546s Starting Epoch 8: Training loss = 117.017708, training accuracy = 0.961405 Evaluation accuracy = 0.973257, Elapsed Time = 1.863692s Starting Epoch 9: Training loss = 110.554947, training accuracy = 0.963241 Evaluation accuracy = 0.974659, Elapsed Time = 1.614416s root@71ac539cda77:~/dcsysh/singa/examples/autograd# mpiexec -np 2 python3 train_mpi.py cnn mnist --op sparseThreshold --spars 0.05 --lr 0.01 Starting Epoch 0: Training loss = 823.523926, training accuracy = 0.701756 Evaluation accuracy = 0.883413, Elapsed Time = 1.945497s Starting Epoch 1: Training loss = 278.808777, training accuracy = 0.906333 Evaluation accuracy = 0.950421, Elapsed Time = 1.600112s Starting Epoch 2: Training loss = 186.442337, training accuracy = 0.937250 Evaluation accuracy = 0.967748, Elapsed Time = 1.615348s Starting Epoch 3: Training loss = 145.515228, training accuracy = 0.951305 Evaluation accuracy = 0.964944, Elapsed Time = 1.599972s Starting Epoch 4: Training loss = 119.955498, training accuracy = 0.959502 Evaluation accuracy = 0.968650, Elapsed Time = 1.601941s Starting Epoch 5: Training loss = 105.754234, training accuracy = 0.965077 Evaluation accuracy = 0.976562, Elapsed Time = 1.601124s Starting Epoch 6: Training loss = 97.148003, training accuracy = 0.967732 Evaluation accuracy = 0.975661, Elapsed Time = 1.600486s Starting Epoch 7: Training loss = 88.760986, training accuracy = 0.969351 Evaluation accuracy = 0.982672, Elapsed Time = 1.597402s Starting Epoch 8: Training loss = 81.591782, training accuracy = 0.973124 Evaluation accuracy = 0.979868, Elapsed Time = 1.584163s Starting Epoch 9: Training loss = 76.670799, training accuracy = 0.974426 Evaluation accuracy = 0.979868, Elapsed Time = 1.564388s ```
---------------------------------------------------------------- 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] With regards, Apache Git Services
