[detailed] use module object id to cache the sharded modules #3591
+255
−6
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
meta-cla
bot
added
the
CLA Signed
Contributor
|
@TroyGarden has exported this pull request. If you are a Meta employee, you can view the originating Diff in D88218200. |
TroyGarden
added a commit
to TroyGarden/torchrec
that referenced
this pull request
Dec 11, 2025
Summary:
# context
* TorchRec relies on `nn.Module.named_children()` to traverse the model to find the sparse modules to shard.
* In a normal case, every sparse module only appears once in the model hiearchy, i.e., it has **only one** parent module.
* However, in some corner cases, a sparse module might have multiple parent modules. This might confuse the TorchRec sharder due to its traversing logic: the same sparse module has multiple FQNs, and hence being sharded multiple times (create multiple sharded modules according to the FQNs).
{F1983896519}
# solution
* cache the sharded module with the original sparse module's object id
* when the sparse module has multiple FQNs, only the first time in the `named_children` traversing, a sharded module will be created.
# changes
* the change is protected by a KillSwitch: [enable_module_id_cache_for_dmp_shard_modules](https://www.internalfb.com/intern/justknobs/?name=pytorch%2Ftorchrec#enable_module_id_cache_for_dmp_shard_modules)
https://fb.workplace.com/groups/429376538334034/permalink/1343336826937996/
Reviewed By: malaybag, iamzainhuda
Differential Revision: D88218200
TroyGarden
force-pushed
the
export-D88218200
branch
2 times, most recently
from
bbbcea6 to
89df447
Compare
TroyGarden
added a commit
to TroyGarden/torchrec
that referenced
this pull request
Dec 11, 2025
Summary:
# context
* TorchRec relies on `nn.Module.named_children()` to traverse the model to find the sparse modules to shard.
* In a normal case, every sparse module only appears once in the model hiearchy, i.e., it has **only one** parent module.
* However, in some corner cases, a sparse module might have multiple parent modules. This might confuse the TorchRec sharder due to its traversing logic: the same sparse module has multiple FQNs, and hence being sharded multiple times (create multiple sharded modules according to the FQNs).
{F1983896519}
# solution
* cache the sharded module with the original sparse module's object id
* when the sparse module has multiple FQNs, only the first time in the `named_children` traversing, a sharded module will be created.
# changes
* the change is protected by a KillSwitch: [enable_module_id_cache_for_dmp_shard_modules](https://www.internalfb.com/intern/justknobs/?name=pytorch%2Ftorchrec#enable_module_id_cache_for_dmp_shard_modules)
https://fb.workplace.com/groups/429376538334034/permalink/1343336826937996/
Reviewed By: malaybag, iamzainhuda
Differential Revision: D88218200
Summary:
# context
* TorchRec relies on `nn.Module.named_children()` to traverse the model to find the sparse modules to shard.
* In a normal case, every sparse module only appears once in the model hiearchy, i.e., it has **only one** parent module.
* However, in some corner cases, a sparse module might have multiple parent modules. This might confuse the TorchRec sharder due to its traversing logic: the same sparse module has multiple FQNs, and hence being sharded multiple times (create multiple sharded modules according to the FQNs).
{F1983896519}
# solution
* cache the sharded module with the original sparse module's object id
* when the sparse module has multiple FQNs, only the first time in the `named_children` traversing, a sharded module will be created.
# changes
* the change is protected by a KillSwitch: [enable_module_id_cache_for_dmp_shard_modules](https://www.internalfb.com/intern/justknobs/?name=pytorch%2Ftorchrec#enable_module_id_cache_for_dmp_shard_modules)
https://fb.workplace.com/groups/429376538334034/permalink/1343336826937996/
Reviewed By: malaybag, iamzainhuda
Differential Revision: D88218200
TroyGarden
force-pushed
the
export-D88218200
branch
from
89df447 to
7a26d0a
Compare
TroyGarden
changed the title
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
context
nn.Module.named_children()to traverse the model to find the sparse modules to shard.solution
named_childrentraversing, a sharded module will be created.changes
https://fb.workplace.com/groups/429376538334034/permalink/1343336826937996/
Differential Revision: D88218200