[FSDP][BE] Move dynamo annotation to separate file #89890
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
[ghstack-poisoned]
awgu
requested review from
H-Huang,
kwen2501,
mrshenli,
pritamdamania87,
rohan-varma and
zhaojuanmao
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/89890
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 7 PendingAs of commit 07c3cd8: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
the
release notes: distributed (fsdp)
awgu
commented
Nov 30, 2022
| ``use_orig_params`` setting passed to the FSDP constructor. | ||
| """ | ||
| for submodule in module.modules(): | ||
| if submodule not in ignored_modules: |
There was a problem hiding this comment.
All of the comments below are copied directly.
This PR makes two minor changes: It (1) moves the recently-added module annotation logic for dynamo support to a separate file `torch/distributed/fsdp/_dynamo_utils.py` and (2) saves the annotated attribute names to global variables `FSDP_MANAGED_MODULE` and `FSDP_USE_ORIG_PARAMS`. - The first change is mainly a personal choice, where I wanted to avoid the dynamo explanation from dominating the FSDP constructor space-wise. I added the `(see function for details)` to the inline comment to forward interested readers. - The second change follows the custom we have taken in the past for such attributes (e.g. `FSDP_FLATTENED`). My understanding (in the past as well as currently) is that this is a good practice. cc @mlazos @soumith @voznesenskym @yanboliang @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @chunyuan-w @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @desertfire [ghstack-poisoned]
This PR makes two minor changes: It (1) moves the recently-added module annotation logic for dynamo support to a separate file `torch/distributed/fsdp/_dynamo_utils.py` and (2) saves the annotated attribute names to global variables `FSDP_MANAGED_MODULE` and `FSDP_USE_ORIG_PARAMS`. - The first change is mainly a personal choice, where I wanted to avoid the dynamo explanation from dominating the FSDP constructor space-wise. I added the `(see function for details)` to the inline comment to forward interested readers. - The second change follows the custom we have taken in the past for such attributes (e.g. `FSDP_FLATTENED`). My understanding (in the past as well as currently) is that this is a good practice. cc @mlazos @soumith @voznesenskym @yanboliang @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @chunyuan-w @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @desertfire [ghstack-poisoned]
This PR makes two minor changes: It (1) moves the recently-added module annotation logic for dynamo support to a separate file `torch/distributed/fsdp/_dynamo_utils.py` and ~~(2) saves the annotated attribute names to global variables `FSDP_MANAGED_MODULE` and `FSDP_USE_ORIG_PARAMS`~~. Update: Since the distributed package may not be included in some builds, it is not safe to import from `torch.distributed...` to a file in `_dynamo/`. I will not include change (2) in this PR. The alternative is to define those globals (privately) in the dynamo file and import from there in the FSDP file. - The first change is mainly a personal choice, where I wanted to avoid the dynamo explanation from dominating the FSDP constructor space-wise. I added the `(see function for details)` to the inline comment to forward interested readers. - The second change follows the custom we have taken in the past for such attributes (e.g. `FSDP_FLATTENED`). My understanding (in the past as well as currently) is that this is a good practice. cc @mlazos @soumith @voznesenskym @yanboliang @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @chunyuan-w @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @desertfire [ghstack-poisoned]
awgu
pushed a commit
to awgu/pytorch
that referenced
this pull request
Nov 30, 2022
ghstack-source-id: 7fd3c5c Pull Request resolved: pytorch#89890
This PR makes two minor changes: It (1) moves the recently-added module annotation logic for dynamo support to a separate file `torch/distributed/fsdp/_dynamo_utils.py` and ~~(2) saves the annotated attribute names to global variables `FSDP_MANAGED_MODULE` and `FSDP_USE_ORIG_PARAMS`~~. Update: Since the distributed package may not be included in some builds, it is not safe to import from `torch.distributed...` to a file in `_dynamo/`. I will not include change (2) in this PR. The alternative is to define those globals (privately) in the dynamo file and import from there in the FSDP file. - The first change is mainly a personal choice, where I wanted to avoid the dynamo explanation from dominating the FSDP constructor space-wise. I added the `(see function for details)` to the inline comment to forward interested readers. - The second change follows the custom we have taken in the past for such attributes (e.g. `FSDP_FLATTENED`). My understanding (in the past as well as currently) is that this is a good practice. cc @mlazos @soumith @voznesenskym @yanboliang @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @chunyuan-w @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @desertfire [ghstack-poisoned]
|
@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 |
kulinseth
pushed a commit
to kulinseth/pytorch
that referenced
this pull request
Dec 10, 2022
This PR makes two minor changes: It (1) moves the recently-added module annotation logic for dynamo support to a separate file `torch/distributed/fsdp/_dynamo_utils.py` and ~~(2) saves the annotated attribute names to global variables `FSDP_MANAGED_MODULE` and `FSDP_USE_ORIG_PARAMS`~~. Update: Since the distributed package may not be included in some builds, it is not safe to import from `torch.distributed...` to a file in `_dynamo/`. I will not include change (2) in this PR. The alternative is to define those globals (privately) in the dynamo file and import from there in the FSDP file. - The first change is mainly a personal choice, where I wanted to avoid the dynamo explanation from dominating the FSDP constructor space-wise. I added the `(see function for details)` to the inline comment to forward interested readers. - The second change follows the custom we have taken in the past for such attributes (e.g. `FSDP_FLATTENED`). My understanding (in the past as well as currently) is that this is a good practice. Pull Request resolved: pytorch#89890 Approved by: https://github.com/wconstab
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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:
This PR makes two minor changes: It (1) moves the recently-added module annotation logic for dynamo support to a separate file
torch/distributed/fsdp/_dynamo_utils.pyand(2) saves the annotated attribute names to global variables.FSDP_MANAGED_MODULEandFSDP_USE_ORIG_PARAMSUpdate: Since the distributed package may not be included in some builds, it is not safe to import from
torch.distributed...to a file in_dynamo/. I will not include change (2) in this PR. The alternative is to define those globals (privately) in the dynamo file and import from there in the FSDP file.(see function for details)to the inline comment to forward interested readers.FSDP_FLATTENED). My understanding (in the past as well as currently) is that this is a good practice.cc @mlazos @soumith @voznesenskym @yanboliang @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @chunyuan-w @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @desertfire