[optim] _actually_ default to foreach #95862
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
This PR is a result of a realization that models are NOT subscribed to the foreach defaulting as have been claimed on our documentation for months now. BIG OOPS. Pull Request resolved: pytorch#95811 Approved by: https://github.com/albanD
Big OOP correction continued. Also added a test this time to verify the defaulting was as expected. The key here is realizing that the grouping for foreach already assumes that the non-param tensorlists follow suit in dtype and device, so it is too narrow to check that _all_ tensors were on CUDA. The main leeway this allowed was state_steps, which are sometimes cpu tensors. Since foreach _can_ handle cpu tensors, this should not introduce breakage. Pull Request resolved: pytorch#95820 Approved by: https://github.com/albanD
janeyx99
requested review from
H-Huang,
albanD,
awgu,
fegin,
kwen2501,
mrshenli,
rohan-varma,
wanchaol and
zhaojuanmao
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/95862
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 3 PendingAs of commit 1bbb43e: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
the
release notes: distributed (fsdp)
pruthvistony
pushed a commit
to ROCm/pytorch
that referenced
this pull request
May 3, 2023
* [optim] include nn.Parameter as foreach supported (pytorch#95811) This PR is a result of a realization that models are NOT subscribed to the foreach defaulting as have been claimed on our documentation for months now. BIG OOPS. Pull Request resolved: pytorch#95811 Approved by: https://github.com/albanD * [optim] Widen the cases for defaulting to foreach (pytorch#95820) Big OOP correction continued. Also added a test this time to verify the defaulting was as expected. The key here is realizing that the grouping for foreach already assumes that the non-param tensorlists follow suit in dtype and device, so it is too narrow to check that _all_ tensors were on CUDA. The main leeway this allowed was state_steps, which are sometimes cpu tensors. Since foreach _can_ handle cpu tensors, this should not introduce breakage. Pull Request resolved: pytorch#95820 Approved by: https://github.com/albanD
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
Big OOP correction. Also added a test this time to verify the defaulting was as expected.
We've claimed + intended to default to foreach as much as we can, but today we realize we missed two critical (the biggest, essentially all real use cases) groups:
Stack from ghstack (oldest at bottom):