DCP HF reader: use safe_open instead of reading the bytes #159406
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
Reading the bytes and converting to tensors is much slower than using safe_open. For a 8B model across 8 ranks, took ~30s to load before this change and ~4s after. Differential Revision: [D78994259](https://our.internmc.facebook.com/intern/diff/D78994259/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/159406
Note: Links to docs will display an error until the docs builds have been completed. ⏳ 1 Pending, 1 Unrelated FailureAs of commit 12b6ef4 with merge base 93da995 ( UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
ankitageorge
added a commit
that referenced
this pull request
Jul 29, 2025
Reading the bytes and converting to tensors is much slower than using safe_open. For a 8B model across 8 ranks, took ~30s to load before this change and ~4s after. Differential Revision: [D78994259](https://our.internmc.facebook.com/intern/diff/D78994259/) ghstack-source-id: 298671089 Pull Request resolved: #159406
|
This pull request was exported from Phabricator. Differential Revision: D78994259 |
Reading the bytes and converting to tensors is much slower than using safe_open. For a 8B model across 8 ranks, took ~30s to load before this change and ~4s after. Differential Revision: [D78994259](https://our.internmc.facebook.com/intern/diff/D78994259/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k pragupta [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Aug 1, 2025
Pull Request resolved: #159406 Reading the bytes and converting to tensors is much slower than using safe_open. For a 8B model across 8 ranks, took ~30s to load before this change and ~4s after. Differential Revision: [D78994259](https://our.internmc.facebook.com/intern/diff/D78994259/) ghstack-source-id: 300113574
|
This pull request was exported from Phabricator. Differential Revision: D78994259 |
Reading the bytes and converting to tensors is much slower than using safe_open. For a 8B model across 8 ranks, took ~30s to load before this change and ~4s after. Differential Revision: [D78994259](https://our.internmc.facebook.com/intern/diff/D78994259/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k pragupta [ghstack-poisoned]
ankitageorge
added a commit
that referenced
this pull request
Aug 1, 2025
Pull Request resolved: #159406 Reading the bytes and converting to tensors is much slower than using safe_open. For a 8B model across 8 ranks, took ~30s to load before this change and ~4s after. ghstack-source-id: 300151112 Differential Revision: [D78994259](https://our.internmc.facebook.com/intern/diff/D78994259/)
|
This pull request was exported from Phabricator. Differential Revision: D78994259 |
saumishr
approved these changes
Aug 1, 2025
Reading the bytes and converting to tensors is much slower than using safe_open. For a 8B model across 8 ranks, took ~30s to load before this change and ~4s after. Differential Revision: [D78994259](https://our.internmc.facebook.com/intern/diff/D78994259/) cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k pragupta [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D78994259 |
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
|
Starting merge as part of PR stack under #159681 |
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, 2025
Get rid of the logic to read the metadata from the header of the safetensors file manually and use the functions as part of safe_open() to get the metadata. This is much cleaner and allows us to not rely on our own custom methods to get metadata, but use safetensors provided APIs Differential Revision: [D79460272](https://our.internmc.facebook.com/intern/diff/D79460272/) Pull Request resolved: #159681 Approved by: https://github.com/saumishr ghstack dependencies: #159405, #159406
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):
Reading the bytes and converting to tensors is much slower than using safe_open. For a 8B model across 8 ranks, took ~30s to load before this change and ~4s after.
Differential Revision: D78994259
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @pragupta