did ```calculate_auto_lr``` affect a lot? about openchat

[imoneoi]

Sorry, what do you mean? This function is for calculating the learning rate based on the batch size using square root rule, as mentioned in the README.

[fpcsong]

Sorry for not making it clear, I wonder if the square root rule has a big impact on the results, compared to other lr scheduler(cosine?)

[imoneoi]

It's using the cosine scheduler. The square root rule is used for calculating initial lr for the schedule, otherwise we need to manually tune it.

[fpcsong]

Thank you~