Where does the model's new weights and parameters get saved after fine tuning?


#1

I followed the step by step the same procedure as the example given here. Is it automatically updated in the json and params files???
Edit:
I got this output btw. :grinning:

(setting env variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable)
2018-07-13 02:09:46,993 Epoch[0] Train-accuracy=0.833333
2018-07-13 02:09:46,993 Epoch[0] Time cost=8.055
2018-07-13 02:09:47,839 Epoch[0] Validation-accuracy=1.000000
2018-07-13 02:09:51,019 Epoch[1] Train-accuracy=1.000000
2018-07-13 02:09:51,019 Epoch[1] Time cost=3.179
2018-07-13 02:09:51,703 Epoch[1] Validation-accuracy=1.000000
2018-07-13 02:09:54,875 Epoch[2] Train-accuracy=1.000000
2018-07-13 02:09:54,875 Epoch[2] Time cost=3.171
2018-07-13 02:09:55,725 Epoch[2] Validation-accuracy=1.000000
2018-07-13 02:09:58,896 Epoch[3] Train-accuracy=1.000000
2018-07-13 02:09:58,896 Epoch[3] Time cost=3.171
2018-07-13 02:09:59,576 Epoch[3] Validation-accuracy=1.000000
2018-07-13 02:10:02,161 Epoch[4] Train-accuracy=1.000000
2018-07-13 02:10:02,162 Epoch[4] Time cost=2.585
2018-07-13 02:10:02,992 Epoch[4] Validation-accuracy=1.000000
2018-07-13 02:10:06,174 Epoch[5] Train-accuracy=1.000000
2018-07-13 02:10:06,174 Epoch[5] Time cost=3.181
2018-07-13 02:10:07,008 Epoch[5] Validation-accuracy=1.000000
2018-07-13 02:10:10,180 Epoch[6] Train-accuracy=1.000000
2018-07-13 02:10:10,180 Epoch[6] Time cost=3.172
2018-07-13 02:10:10,860 Epoch[6] Validation-accuracy=1.000000
2018-07-13 02:10:14,032 Epoch[7] Train-accuracy=1.000000
2018-07-13 02:10:14,032 Epoch[7] Time cost=3.172
2018-07-13 02:10:14,870 Epoch[7] Validation-accuracy=1.000000


#2

Hi @anshumankmr,

So the parameters are updated in memory and not in the params file. Unless you specifically call the save_checkpoint method of your Module, or use mxnet.callback.module_checkpoint as the epoch_end_callback in your Module to save the parameters to file at the end of each epoch.