gluon-nlp/scripts/question_answering/train.py

[rrjia]

Description

class ExponentialMovingAverage
in train.py the ema use two position:

if global_step == 1:
for name, param in model.collect_params().items():
ema.add(name, param.data(CTX[0]))

for name, param in model.collect_params().items():
ema(name, param.data(CTX[0]))

but model‘s params not update ,final we save model params as:

save model after train one epoch

    model.save_parameters(SAVE_MODEL_PREFIX_NAME +
                          time.asctime(time.localtime(time.time())))

is the ema not work?

[Leonard Lausen]

There is no train.py in https://github.com/dmlc/gluon-nlp/tree/master/scripts/question_answering or https://github.com/dmlc/gluon-nlp/tree/v0.x/scripts/question_answering

Please provide a link to the code you're concerned about. Thanks

[rrjia]

code link is : https://github.com/dmlc/gluon-nlp/blob/qanet/scripts/question_answering/train.py

appreciate for your supprot.