[DeviceMesh] Update slicing documentation to include nD and non-continuous slicing #132311
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/132311
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (1 Unrelated Failure)As of commit 9ff80df with merge base 3864a2d ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
the
oncall: distributed
wz337
changed the title
wz337
added
module: dtensor
wz337
added a commit
to wz337/pytorch
that referenced
this pull request
Aug 1, 2024
ghstack-source-id: a9014ef Pull Request resolved: pytorch#132311
wanchaol
requested changes
Aug 2, 2024
torch/distributed/device_mesh.py
Outdated
| Calling mesh["dp"] on rank 3, 7 would return a 1D child DeviceMesh:([3, 7]). | ||
| The following program runs on each process/rank in an SPMD manner in a world size of 8. | ||
| In the first example: | ||
| Calling mesh_2d["tp"] on rank 0, 1, 2, 3 returns a 1D child DeviceMesh:([0, 1, 2, 3]). |
There was a problem hiding this comment.
I think we should change the wording from "child DeviceMesh" to sth like "sub DeviceMesh"? A child indicates a parent somewhere, where we should avoid using this term going forward. Please rename all the comments
There was a problem hiding this comment.
Removed "child DeviceMesh" and replaced with submesh instead.
I am not sure if we want to expose the concept of root mesh yet, as root mesh only lives in our mesh env right now. I think, we should leave out the term 'root mesh` from public API documentation.
[Edit]: If later we decided to have a public API to let users know whether a mesh is a root mesh, then we can include the term 'root_mesh' in the documentation.
torch/distributed/device_mesh.py
Outdated
| """ | ||
| Slice the current DeviceMesh based on the mesh_dim_name given to create a child | ||
| DeviceMesh. | ||
| Slice the current DeviceMesh based on the mesh_dim_names given to create a child |
There was a problem hiding this comment.
Suggested change
| Slice the current DeviceMesh based on the mesh_dim_names given to create a child | |
| Derive a sub-DeviceMesh based on the root DeviceMesh based on the ``mesh_dim_names`` provided |
torch/distributed/device_mesh.py
Outdated
| Slice the current DeviceMesh based on the mesh_dim_name given to create a child | ||
| DeviceMesh. | ||
| Slice the current DeviceMesh based on the mesh_dim_names given to create a child | ||
| DeviceMesh. This supports 1D or nD slicing. The DeviceMesh created will have dimensions |
There was a problem hiding this comment.
Suggested change
| DeviceMesh. This supports 1D or nD slicing. The DeviceMesh created will have dimensions | |
| The DeviceMesh created consists of the dimensions and the communicators indicated by ``mesh_dim_names`` |
wanchaol
approved these changes
Aug 5, 2024
|
@pytorchmergebot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
pytorchmergebot
pushed a commit
that referenced
this pull request
Aug 7, 2024
…root mesh (#132339) Previously, when we slice out a submesh from a mesh, we assign the mesh as the parent mesh of the submesh. In this case, when we have a 3D mesh topology, the parent mesh of a 1D mesh sliced out from the 3D mesh is different from the parent mesh of the same 1D mesh sliced out from the 2D submesh of the 3D mesh. For example: ``` mesh_3d = init_device_mesh("cuda", (2,2,2), ("dim0", "dim1", "dim2")) mesh_dim0 = mesh_3d["dim0"] mesh_2d = mesh_2d["dim0", "dim1"] mesh_dim0_2 = mesh_2d["dim0_2"] # This would evaluate to be True print(_mesh_resources.get_parent_mesh(mesh_dim0) != _mesh_resources.get_parent_mesh(mesh_dim0)) ``` We can always reconstruct the mesh needed from the mesh dim names, as long as two dims come from the same root. For simplicity, we do not see the necessity of building a tree structure to represent child-parent relationship. Therefore, we are replacing the parent mesh concept with a root mesh concept in `_MeshEnv` so we would have: ``` mesh_3d = init_device_mesh("cuda", (2,2,2), ("dim0", "dim1", "dim2")) mesh_dim0 = mesh_3d["dim0"] mesh_2d = mesh_2d["dim0", "dim1"] mesh_dim0_2 = mesh_2d["dim0_2"] # This would evaluate to be True print(_mesh_resources.get_root_mesh(mesh_dim0) == _mesh_resources.get_root_mesh(mesh_dim0)) ``` With this change, we will have two types of meshes in an environment. 1. `device_mesh != _mesh_resources.get_root_mesh(device_mesh)` means that the device_mesh is created by slicing. 2. `device_mesh == _mesh_resources.get_root_mesh(device_mesh)` means that the device_mesh is a root mesh not created through slicing. Pull Request resolved: #132339 Approved by: https://github.com/wanchaol ghstack dependencies: #132310, #132311
pytorchmergebot
pushed a commit
that referenced
this pull request
Aug 8, 2024
Adds a new private API to flatten a DeviceMesh to a 1D DeviceMesh such that:
```
mesh_3d = init_device_mesh(
self.device_type, (2, 2, 2), mesh_dim_names=("dp", "cp", "tp"),
)
dp_cp_mesh = mesh_3d["dp", "cp"]
# flattened_mesh on rank 0, 2, 4, 6 is DeviceMesh([0, 2, 4, 6], mesh_dim_names=('dp_cp',))
# flattened_mesh on rank 1, 3, 5, 7 is DeviceMesh([1, 3, 5, 7], mesh_dim_names=('dp_cp',))
flattened_dp_cp_mesh = dp_cp_mesh._flatten()
```
Pull Request resolved: #132632
Approved by: https://github.com/fegin, https://github.com/wanchaol
ghstack dependencies: #132310, #132311, #132339
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Stack from ghstack (oldest at bottom):
cc @XilunWu @H-Huang @awgu @kwen2501 @wanchaol @fegin @fduwjj @wconstab @d4l3k @c-p-i-o @tianyu-l @chauhang