Adaptive learning rate is a thing. For example, one scheme I've used before is to decrease the learning rate if the validation loss stops decreasing. It's not

Adaptive learning rate is a thing. For example, one scheme I've used before is to decrease the learning rate if the validation loss stops decreasing. It's not clear to me if this is applicable to LLMs though.