Question about the training_reuse_numb_steps

[xiongluojie]

When I continue training a model from the last iteration, what is the appropriate value for training_reuse_numb_steps? Can it be significantly fewer than the number of steps used in the initial training of the model?
Thanks!

[wanghan-iapcm]

may not always be true. the new data (data labeled by the previous iteration) should be properly trained with training_reuse_numb_steps training steps.

[xiongluojie]

I see. Thank you for your help!