AdamW operator (Fixing Weight Decay Regularization in Adam)#13728
AdamW operator (Fixing Weight Decay Regularization in Adam)#13728eric-haibin-lin merged 3 commits intoapache:masterfrom
Conversation
|
@sxjscience @szhengac could you guys help review this PR? |
eric-haibin-lin
changed the title
python/mxnet/optimizer/optimizer.py
Outdated
| rescaled_grad = clip(grad * rescale_grad, clip_gradient) | ||
| m = beta1 * m + (1 - beta1) * rescaled_grad | ||
| v = beta2 * v + (1 - beta2) * (rescaled_grad**2) | ||
| w = w - learning_rate * (m / (sqrt(v) + epsilon) + wd * w) |
There was a problem hiding this comment.
According to the paper, it has two learning rates. An alpha before m / (sqrt(v) + epsilon).
There was a problem hiding this comment.
Good point. The issue is that the learning rate and schedule multiplier is not decoupled in MXNet. Here learning_rate is effectively eta_t * alpha in the paper and wd actually needs to be set as w / alpha. In another word wd can be rescaled properly so that it does exactly the same thing in the paper. Would this be acceptable? Is so maybe I can move this to contrib for the moment
There was a problem hiding this comment.
I think it's acceptable as long as the wd is set correctly.
There was a problem hiding this comment.
On second thought I think it's better to keep it consistent with the paper
sandeep-krishnamurthy
left a comment
sandeep-krishnamurthy
left a comment
There was a problem hiding this comment.
Thanks.
Can you please provide an output of an end to end use case using AdamW optimizer?
sandeep-krishnamurthy
added
Optimizer
pr-awaiting-review
|
@sandeep-krishnamurthy training/fine-tuning the BERT model in GluonNLP would be a use case of AdamW |
python/mxnet/optimizer/optimizer.py
Outdated
| kwargs['clip_gradient'] = self.clip_gradient | ||
|
|
||
| mean, var = state | ||
| adamw_update(weight, grad, mean, var, out=weight, lr=lr, wd=wd, **kwargs) |
There was a problem hiding this comment.
Should we set wd to something like wd / self._original_lr?
|
@sxjscience @szhengac I took a step back and moved the operator to contrib and use the same notation as the one in the paper. I think the optimizer API still needs more discussion, so I removed it from the PR. |
eric-haibin-lin
changed the title
…3728) * tests * remove optimizer and move op to contrib * rename parameter
…3728) * tests * remove optimizer and move op to contrib * rename parameter
4 participants
Description
Implement a modification of Adam in "Fixing Weight Decay Regularization in Adam" https://arxiv.org/abs/1711.05101.
Checklist
Essentials
Please feel free to remove inapplicable items for your PR.