How should I set the optimizer parameter in train.yaml
fangg2021 opened this issue · comments
Who can explain the optimizer parameter in train.yaml:
batch_ size: 16
betas: [0.9, 0.98]
eps: 0.000000001
weight_ decay: 0.0
grad_ clip_ thresh: 1.0
grad_ acc_ step: 1
warm_ up_ step: 4000
anneal_ steps: [300000, 400000, 500000]
anneal_ rate: 0.3
How should I set it? For example, my samples are only 3000. Thank you.
one more question, How to change the time occupied by 'sp'? Or where?