When setting override-opt_param-scheduler (but still load optimizer and load rng) and setting new learning rate scheduler (including max lr, min lr, decay style, etc.), the learning rate still persists its original scheduler.
A related issue could be #963
To Reproduce
Set max_lr as 6e-4 and constant learning rate.
Training some steps and save checkpoint.
Load the checkpoint (including optimizer params) and override the scheduler (e.g. cosine 6e-4 to 6e-5).
Then the bug shows: the learning rate is still constantly 6e-4.
Expected behavior
The learning rate scheduler should be overrided.
Environment (please complete the following information):
Megatron-LM commit ID [9bcd417]
Proposed fix
The following code (https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/core/optimizer_param_scheduler.py#L121) leads to reloading the optimizer's learning rate when setting override-opt_param-scheduler
def get_lr(self, param_group: dict) -> float:
"""Learning rate decay functions from:
https://openreview.net/pdf?id=BJYwwY9ll pg. 4
Args:
param_group (dict): parameter group from the optimizer.
"""
max_lr = param_group.get('max_lr', self.max_lr)
min_lr = param_group.get('min_lr', self.min_lr)
A possible solution could be
def get_lr(self, param_group: dict) -> float:
"""Learning rate decay functions from:
https://openreview.net/pdf?id=BJYwwY9ll pg. 4
Args:
param_group (dict): parameter group from the optimizer.
When setting override-opt_param-scheduler (but still load optimizer and load rng) and setting new learning rate scheduler (including max lr, min lr, decay style, etc.), the learning rate still persists its original scheduler.
A related issue could be #963
To Reproduce
Set max_lr as 6e-4 and constant learning rate.
Training some steps and save checkpoint.
Load the checkpoint (including optimizer params) and override the scheduler (e.g. cosine 6e-4 to 6e-5).
Then the bug shows: the learning rate is still constantly 6e-4.
Expected behavior
The learning rate scheduler should be overrided.
Environment (please complete the following information):
Megatron-LM commit ID [9bcd417]
Proposed fix
The following code (https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/core/optimizer_param_scheduler.py#L121) leads to reloading the optimizer's learning rate when setting override-opt_param-scheduler
def get_lr(self, param_group: dict) -> float:
"""Learning rate decay functions from:
https://openreview.net/pdf?id=BJYwwY9ll pg. 4
Args:
param_group (dict): parameter group from the optimizer.
"""
max_lr = param_group.get('max_lr', self.max_lr)
min_lr = param_group.get('min_lr', self.min_lr)
A possible solution could be
def get_lr(self, param_group: dict) -> float:
"""Learning rate decay functions from:
https://openreview.net/pdf?id=BJYwwY9ll pg. 4
Args:
param_group (dict): parameter group from the optimizer.
