You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: _posts/2023-07-31-performant-distributed-checkpointing.md
+1-1
Original file line number
Diff line number
Diff line change
@@ -31,7 +31,7 @@ A major improvement from 1.13 to 2.0 includes adding sharded_state_dict support
31
31
32
32
Another major change is that we decouple the storage layer from the checkpoint planning layer and separate implementation from the interface for both layers. With this change, users can now specify how their state_dict should be chunked or transformed during the checkpoint planning phase. Additionally, the customizable storage layer can easily accommodate different backends.
33
33
34
-
More information on the Distributed Checkpoint package can be found [here](https://fb.workplace.com/notes/1247127459498709).
34
+
More information on the Distributed Checkpoint package can be found [here](https://pytorch.org/docs/stable/distributed.checkpoint.html).
35
35
36
36
## Performant Distributed checkpointing in Production with IBM
0 commit comments