Script for consolidation of sharded safetensor files #154743
Closed
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
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) [ghstack-poisoned]
This was referenced May 30, 2025
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/154743
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 1 PendingAs of commit fad5fa1 with merge base f79689b ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
oncall: distributed
ankitageorge
added a commit
that referenced
this pull request
May 30, 2025
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) ghstack-source-id: 287228528 Pull Request resolved: #154743
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
May 30, 2025
Pull Request resolved: #154743 Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) ghstack-source-id: 287291639
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
This was referenced Jun 11, 2025
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Jun 13, 2025
Pull Request resolved: #154743 Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. ghstack-source-id: 290286966 Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/)
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Jun 13, 2025
Pull Request resolved: #154743 Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. ghstack-source-id: 290311030 Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/)
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
This was referenced Jun 13, 2025
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Jun 23, 2025
Pull Request resolved: #154743 Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. ghstack-source-id: 292189152 Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/)
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Jun 23, 2025
Pull Request resolved: #154743 Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. ghstack-source-id: 292214509 Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/)
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Jun 24, 2025
Pull Request resolved: #154743 Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. ghstack-source-id: 292257534 Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/)
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
saumishr
approved these changes
Jun 30, 2025
| num_threads: Number of threads to use for parallel processing of saving data to output files. | ||
| """ | ||
| # Create filesystem using fsspec for file operations | ||
| input_fs, _ = url_to_fs(input_dir) |
There was a problem hiding this comment.
Same comment here as well. Lets add some timing logs to keep track of the performance of this script.
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Jun 30, 2025
Pull Request resolved: #154743 Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. ghstack-source-id: 293486705 Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/)
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Jun 30, 2025
Pull Request resolved: #154743 Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory. ghstack-source-id: 293495694 Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/)
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
|
@pytorchmergebot merge -i |
Merge startedYour change will be merged while ignoring the following 1 checks: trunk / verify-cachebench-cpu-build / build 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
Jul 1, 2025
…h step (#156705) Title - we can consolidate the shards to a full tensors, optionally behind a flag, in the finish step of DCP.save also adds the thread count argument which is configurable for users, before we were just using the default of 1. Re-creating #155940 bc it got into a bad detached state Differential Revision: [D77231774](https://our.internmc.facebook.com/intern/diff/D77231774/) Pull Request resolved: #156705 Approved by: https://github.com/saumishr ghstack dependencies: #154743
|
This pull request was exported from Phabricator. Differential Revision: D75536985 |
pytorchmergebot
pushed a commit
that referenced
this pull request
Jul 2, 2025
Need to change an argument name that was changed in the test so that it doesn't throw Differential Revision: [D77604210](https://our.internmc.facebook.com/intern/diff/D77604210/) Pull Request resolved: #157386 Approved by: https://github.com/meetv18 ghstack dependencies: #154743, #156705
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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):
Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory.
Differential Revision: D75536985
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k