--restart run with a new learning rate?

[Lewis-YL]

Hi guys,

Is there a way to reset the index of the checkpoint when using --restart?

According to the manual,
lr(t) = start_lr * decay_rate ^ ( t / decay_steps )

When t is 1 million steps or even larger,
it is hard to adjust the rate at the checkpoint based on start_lr and decay_steps.

[Lewis-YL]

Problem solved.

[lnnbig]

Hi, may I ask how to restart with a new learning rate? Many thanks!