Skip to content

Support of dtensor redistribute with device order #160266

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 3 commits into
base: gh/zpcore/12/base
Choose a base branch
from
+317 −74
Open

Support of dtensor redistribute with device order #160266

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e99b775
Support of dtensor redistribute with device order
zpcore Aug 10, 2025
0ca96f6
Update on "Support of dtensor redistribute with device order"
zpcore Aug 10, 2025
12fe67c
Update on "Support of dtensor redistribute with device order"
zpcore Aug 10, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Update on "Support of dtensor redistribute with device order" 
[Prototype; for RFC, not ready for review]

Now redistribute dtensor honors the device ordering. If no order information specified, it will use the default device order [0,1,2,...]. We can specify `device_order` as follow:
```
sharded_dt = distribute_tensor(input_data, mesh, placement, device_order)
```
and
```
out_dt = sharded_dt.redistribute(mesh, placement, device_order)
```

Note that device order information is added into the DTensorSpec. So `redistribute_local_tensor` doesn't need the `src_device_order` and `dst_device_order`. I leave them here as a reference for AutoParallel (cc fmassa ). I will remove those order related args from redistributed related API in this PR.





cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k pragupta

[ghstack-poisoned]
  • Loading branch information
@zpcore
zpcore committed Aug 10, 2025
commit 12fe67c7ec73dfb1a8ce1f082638af812ef5bdcf
2 changes: 1 addition & 1 deletion torch/distributed/tensor/_redistribute.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -193,7 +193,7 @@ def _map_tensor_dim_to_mesh_dim(placements, device_order):
and dst_device_order_to_mesh_dims[j] == [mesh_dim]
):
mesh_dim_size = device_mesh.size(mesh_dim=mesh_dim)
current_placement = sorted_dst_placement[mesh_dim] # <<<<<<<<<<<<<<<<<<<<<<, error
current_placement = sorted_dst_placement[mesh_dim]
assert isinstance(current_placement, Shard)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I might be doing something wrong when trying this out, but I'm hitting this assertion, where current_placement is Replicate.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you have instructions on how to reproduce the issue? I tried python examples/example_autoparallel.py and it complained RuntimeError: Function CompiledFunctionBackward returned an invalid gradient at index 0 - got [1, 6144] but expected shape compatible with [24, 6144].

# alltoall from Shard(tensor_dim) to Shard()
transform_infos.append(
Expand Down
2 changes: 1 addition & 1 deletion torch/distributed/tensor/placement_types.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,7 +121,7 @@ def _local_shard_size_and_offset(
computes the new local shard size and offset given the desired number of chunks
(num_chunks is generally equal to the size of the current sharding dim).

Note: T234040481 new local shard offset is relative to the current sharded tensor, not the global tensor.
Note: new local shard offset is relative to the current sharded tensor, not the global tensor.
See `_utils.compute_local_shape_and_global_offset` for computing global offset.

Returns (new local shard size, offset)
Expand Down
Loading