saving model when eval_period and checkpoint_period are not sync

[ghazni123]

I am using tools/plain_train_net.py to train model. training logic is exactly same do_train(). My settings are:

cfg.TEST.EVAL_PERIOD = 29544
cfg.SOLVER.IMS_PER_BATCH = 6
cfg.SOLVER.CHECKPOINT_PERIOD = 5000
cfg.SOLVER.BASE_LR = 0.0025
cfg.SOLVER.MAX_ITER = 1000000
cfg.SOLVER.STEPS = (500000, 750000)

1: Is there a way that i can tell the training logic to save the checkpoint when evaluation is done? for the case when eval_period and checkpoint_period are not aligned as in specifications above.

2: How do i save additional elements in checkpoint?

Thank you.