[SymmMem] Speed up tests #153677
Closed
[SymmMem] Speed up tests #153677
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/153677
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit c5d9c3e with merge base fa85434 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
oncall: distributed
fegin
approved these changes
May 16, 2025
Skylion007
approved these changes
May 16, 2025
ngimel
approved these changes
May 16, 2025
| [4, 8192, 8196], | ||
| [4, 8, 16], | ||
| [ | ||
| 8 |
There was a problem hiding this comment.
interetsint, do you know why memory usage changed? All these tests use very little memory
There was a problem hiding this comment.
It it not a problem of the alignment, but the number of tests we run continuously. It seems we either failed to release tensors or there is some flaw in the allocation logic (e.g. allocated more than needed).
|
@pytorchbot 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 |
Merge failedReason: 1 mandatory check(s) failed. The first few are: Dig deeper by viewing the failures on hud |
|
@pytorchbot 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 |
|
@pytorchbot revert -m "I don't know how, but you PRs keep escaping TD and breaking trunk oops I wrong" -c nosignal |
|
Sorry, looks like infra is just unhappy |
|
@pytorchbot successfully started a revert job. Check the current status here. |
|
Don't want to revert based on edited command |
pytorchmergebot
pushed a commit
that referenced
this pull request
Jun 6, 2025
A 2D AllToAllv shuffle is illustrated below:
(`world_size` = 2, `ne` = 2, where `ne` is number of experts per rank)
```
Source: | Rank 0 | Rank 1 |
| c0 | c1 | c2 | c3 | d0 | d1 | d2 | d3 |
Dest : | Rank 0 | Rank 1 |
| c0 | d0 | c1 | d1 | c2 | d2 | c3 | d3 |
```
where each `c_i` / `d_i` are slices of the `input` tensor, targeting expert `i`, with length indicated by input splits (in `in_out_splits[0]`).
That is, the 2D AllToAllv shuffle achieves a transpose from rank-major order at input to expert-major order at output.
Pull Request resolved: #155058
Approved by: https://github.com/ngimel
ghstack dependencies: #153653, #153677
pytorchmergebot
pushed a commit
that referenced
this pull request
Jun 6, 2025
Downstream consumer of the 2D all-to-all-v is often a group GEMM. Today the GEMM often have an alignment requirement on the chunk sizes within grouped sequence, where each chunk carries the tokens headed for an expert. For example, `torch._group_mm` requires an alignment of 8. This PR adds that alignment capability, when user passes in a `major_align` argument, so that no extra padding step is needed. The key in supporting that is making the output offsets aligned to such value. (Output offsets are returned to the users in the 3rd row of `in_out_splits`, on device. The 2nd row, output splits, are unaffected by this alignment value -- i.e. reflecting true number of tokens for an expert.) The algorithm is as follows.  In detailed implementation, we use warp scan to calculate prefix sum on the "block" illustrated above. As a result, the "block" size, i.e. `npes` is currently limited to warp size 32. Pull Request resolved: #155172 Approved by: https://github.com/ngimel ghstack dependencies: #153653, #153677, #155058
vijayabhaskar-ev
pushed a commit
to vijayabhaskar-ev/pytorch
that referenced
this pull request
Jun 22, 2025
A 2D AllToAllv shuffle is illustrated below:
(`world_size` = 2, `ne` = 2, where `ne` is number of experts per rank)
```
Source: | Rank 0 | Rank 1 |
| c0 | c1 | c2 | c3 | d0 | d1 | d2 | d3 |
Dest : | Rank 0 | Rank 1 |
| c0 | d0 | c1 | d1 | c2 | d2 | c3 | d3 |
```
where each `c_i` / `d_i` are slices of the `input` tensor, targeting expert `i`, with length indicated by input splits (in `in_out_splits[0]`).
That is, the 2D AllToAllv shuffle achieves a transpose from rank-major order at input to expert-major order at output.
Pull Request resolved: pytorch#155058
Approved by: https://github.com/ngimel
ghstack dependencies: pytorch#153653, pytorch#153677
vijayabhaskar-ev
pushed a commit
to vijayabhaskar-ev/pytorch
that referenced
this pull request
Jun 22, 2025
…155172) Downstream consumer of the 2D all-to-all-v is often a group GEMM. Today the GEMM often have an alignment requirement on the chunk sizes within grouped sequence, where each chunk carries the tokens headed for an expert. For example, `torch._group_mm` requires an alignment of 8. This PR adds that alignment capability, when user passes in a `major_align` argument, so that no extra padding step is needed. The key in supporting that is making the output offsets aligned to such value. (Output offsets are returned to the users in the 3rd row of `in_out_splits`, on device. The 2nd row, output splits, are unaffected by this alignment value -- i.e. reflecting true number of tokens for an expert.) The algorithm is as follows.  In detailed implementation, we use warp scan to calculate prefix sum on the "block" illustrated above. As a result, the "block" size, i.e. `npes` is currently limited to warp size 32. Pull Request resolved: pytorch#155172 Approved by: https://github.com/ngimel ghstack dependencies: pytorch#153653, pytorch#153677, pytorch#155058
vijayabhaskar-ev
pushed a commit
to vijayabhaskar-ev/pytorch
that referenced
this pull request
Jul 14, 2025
A 2D AllToAllv shuffle is illustrated below:
(`world_size` = 2, `ne` = 2, where `ne` is number of experts per rank)
```
Source: | Rank 0 | Rank 1 |
| c0 | c1 | c2 | c3 | d0 | d1 | d2 | d3 |
Dest : | Rank 0 | Rank 1 |
| c0 | d0 | c1 | d1 | c2 | d2 | c3 | d3 |
```
where each `c_i` / `d_i` are slices of the `input` tensor, targeting expert `i`, with length indicated by input splits (in `in_out_splits[0]`).
That is, the 2D AllToAllv shuffle achieves a transpose from rank-major order at input to expert-major order at output.
Pull Request resolved: pytorch#155058
Approved by: https://github.com/ngimel
ghstack dependencies: pytorch#153653, pytorch#153677
vijayabhaskar-ev
pushed a commit
to vijayabhaskar-ev/pytorch
that referenced
this pull request
Jul 14, 2025
…155172) Downstream consumer of the 2D all-to-all-v is often a group GEMM. Today the GEMM often have an alignment requirement on the chunk sizes within grouped sequence, where each chunk carries the tokens headed for an expert. For example, `torch._group_mm` requires an alignment of 8. This PR adds that alignment capability, when user passes in a `major_align` argument, so that no extra padding step is needed. The key in supporting that is making the output offsets aligned to such value. (Output offsets are returned to the users in the 3rd row of `in_out_splits`, on device. The 2nd row, output splits, are unaffected by this alignment value -- i.e. reflecting true number of tokens for an expert.) The algorithm is as follows.  In detailed implementation, we use warp scan to calculate prefix sum on the "block" illustrated above. As a result, the "block" size, i.e. `npes` is currently limited to warp size 32. Pull Request resolved: pytorch#155172 Approved by: https://github.com/ngimel ghstack dependencies: pytorch#153653, pytorch#153677, pytorch#155058
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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):
Use
MultiProcContinousTestto avoid re-create ProcessGroup in each test instance.cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k