About adap_priors

[dzj441]

nice work, but i met 1 minor puzzles when reading the code: how is adap_priors in ./datasets/adap_priors.py determined ?
These hyper-params seems to influence ilp greatly since they are determining the average tokens allocated for the block in batch ,but no descriptions about selecting them is mentioned in code or paper. I am quite curious about it.