Command line argument of learning rate scheduler

[hlh981029]

#120

[CLAassistant]

All committers have signed the CLA.

[hlh981029]

EasyNLP/easynlp/core/trainer.py

Lines 325 to 334 in 169bf0a

	if self.args.use_amp:
	self._scaler.step(self._optimizer)
	self._scaler.update()
	else:
	if self._lr_scheduler:
	# If use AdamW, it should explicit perform self._optimizer.step()
	# If not, it means use bertadam, which consists of inherent schedular.
	self._optimizer.step()
	if self._lr_scheduler:
	self._lr_scheduler.step()

这里的逻辑我不太理解，比如，self._optimizer 是 BertAdam 的时候 self._lr_scheduler 为 None，这里是需要调用 self._optimizer.step() 的。