-
Notifications
You must be signed in to change notification settings - Fork 1.8k
[Retiarii] cross-graph optimization: device placement and input deduplication #3202
Conversation
learning_rate=learning_rate, weight_decay=weight_decay, optimizer=optimizer) | ||
|
||
|
||
class Classification(Lightning): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is it possible to inherit Classification
and CgoMixin
, to avoid copy all the code.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It seems to be hard. Most lines of code of Classification
and Regression
are documentation. The two lines of code in __init__
are both customized for CGO.
nni/retiarii/experiment/pytorch.py
Outdated
@@ -3,6 +3,7 @@ | |||
|
|||
import atexit | |||
import logging | |||
from nni.retiarii.execution.logical_optimizer.logical_plan import PhysicalDevice |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Suggest add shortcut to simplify import
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It seems that most changes are self-contained and doesn't overlap with other modules. I suggest we merge this PR (as it's large enough) and we can improve later.
No description provided.