Feature distributed fused adam #184
Merged
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
…dependencies - fused adam, distributed adam. Updated the unit test case for distributed fused adam.
…uted fused adam. Skipped these particular UTs
amd-sriram
added a commit
to ROCm/pytorch
that referenced
this pull request
Mar 19, 2025
…locator. This is impacting Distributed Fused Adam in Rocm/APEX. See PR ROCm/apex#184
amd-sriram
added a commit
to ROCm/pytorch
that referenced
this pull request
Mar 20, 2025
…cator. This is impacting Distributed Fused Adam in Rocm/APEX. See PR ROCm/apex#184
pruthvistony
approved these changes
Mar 21, 2025
|
@amd-sriram , |
|
Please cherry-pick this into release/1.6.0 branch. |
@pruthvistony : Yes There is one PR in Pytorch that will impact one UT and the nccl_ub feature in distributed_fused_adam. Kindly review this as well. |
amd-sriram
added a commit
that referenced
this pull request
Mar 24, 2025
* Updated feature of distributed fused adam from upstream. Updated its dependencies - fused adam, distributed adam. Updated the unit test case for distributed fused adam. * Raise Exception when nccl user buffer / cuda graph is used in distributed fused adam. Skipped these particular UTs * Adding support for rccl_ub in distributed_fused_adam * build nccl_allocator module when cuda_ext flag is mentioned
pruthvistony
pushed a commit
that referenced
this pull request
Mar 24, 2025
* Feature distributed fused adam (#184) * Updated feature of distributed fused adam from upstream. Updated its dependencies - fused adam, distributed adam. Updated the unit test case for distributed fused adam. * Raise Exception when nccl user buffer / cuda graph is used in distributed fused adam. Skipped these particular UTs * Adding support for rccl_ub in distributed_fused_adam * build nccl_allocator module when cuda_ext flag is mentioned * Ported distributed fused lamb from upstream repo. Add support for parameters - fused_norm, full_ar, set_param_views_to_flat_buffer, skip_allgather, fuse_scale, param_order, Nccl_allgather_channels (#185) * for distributed fused adam, add condition to remove nccl_allocator only if it is mentioned explicitly
pruthvistony
pushed a commit
to ROCm/pytorch
that referenced
this pull request
Mar 24, 2025
Altering the flag to use the correct streamType for CUDAPluggableAllocator. This is impacting Distributed Fused Adam in Rocm/APEX. See PR ROCm/apex#184 Related Issue : https://ontrack-internal.amd.com/browse/SWDEV-519796
amd-sriram
added a commit
that referenced
this pull request
Mar 25, 2025
* Updated feature of distributed fused adam from upstream. Updated its dependencies - fused adam, distributed adam. Updated the unit test case for distributed fused adam. * Raise Exception when nccl user buffer / cuda graph is used in distributed fused adam. Skipped these particular UTs * Adding support for rccl_ub in distributed_fused_adam * build nccl_allocator module when cuda_ext flag is mentioned
amd-sriram
added a commit
to amd-sriram/pytorch
that referenced
this pull request
Mar 26, 2025
) Altering the flag to use the correct streamType for CUDAPluggableAllocator. This is impacting Distributed Fused Adam in Rocm/APEX. See PR ROCm/apex#184 Related Issue : https://ontrack-internal.amd.com/browse/SWDEV-519796
pytorchmergebot
pushed a commit
to pytorch/pytorch
that referenced
this pull request
Apr 1, 2025
Altering the flag to use the correct streamType in CUDAPluggableAllocator class for ROCm gpu. The flag TORCH_HIP_VERSION does not work for ROCm as intended. This flag is replaced with USE_ROCM. This is impacting Distributed Fused Adam in Rocm/APEX when using nccl_ub feature. This has been tested with rocm/apex. See PR ROCm/apex#184 Pull Request resolved: #150010 Approved by: https://github.com/jeffdaily
amathewc
pushed a commit
to amathewc/pytorch
that referenced
this pull request
Apr 17, 2025
Altering the flag to use the correct streamType in CUDAPluggableAllocator class for ROCm gpu. The flag TORCH_HIP_VERSION does not work for ROCm as intended. This flag is replaced with USE_ROCM. This is impacting Distributed Fused Adam in Rocm/APEX when using nccl_ub feature. This has been tested with rocm/apex. See PR ROCm/apex#184 Pull Request resolved: pytorch#150010 Approved by: https://github.com/jeffdaily
amd-sriram
added a commit
to amd-sriram/pytorch
that referenced
this pull request
Apr 25, 2025
) Altering the flag to use the correct streamType for CUDAPluggableAllocator. This is impacting Distributed Fused Adam in Rocm/APEX. See PR ROCm/apex#184 Related Issue : https://ontrack-internal.amd.com/browse/SWDEV-519796
pytorchbot
pushed a commit
to pytorch/pytorch
that referenced
this pull request
May 20, 2025
Altering the flag to use the correct streamType in CUDAPluggableAllocator class for ROCm gpu. The flag TORCH_HIP_VERSION does not work for ROCm as intended. This flag is replaced with USE_ROCM. This is impacting Distributed Fused Adam in Rocm/APEX when using nccl_ub feature. This has been tested with rocm/apex. See PR ROCm/apex#184 Pull Request resolved: #150010 Approved by: https://github.com/jeffdaily (cherry picked from commit a19b667)
atalman
pushed a commit
to pytorch/pytorch
that referenced
this pull request
May 22, 2025
[ROCm] Update CUDAPluggableAllocator.h (#1984) (#150010) Altering the flag to use the correct streamType in CUDAPluggableAllocator class for ROCm gpu. The flag TORCH_HIP_VERSION does not work for ROCm as intended. This flag is replaced with USE_ROCM. This is impacting Distributed Fused Adam in Rocm/APEX when using nccl_ub feature. This has been tested with rocm/apex. See PR ROCm/apex#184 Pull Request resolved: #150010 Approved by: https://github.com/jeffdaily (cherry picked from commit a19b667) Co-authored-by: Sriram Kumar <[email protected]>
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.
Ported Distributed_fused_adam from Nvidia/apex.
Ref: https://ontrack-internal.amd.com/browse/SWDEV-519796