Avoid graph breaks in torch.compile caused by inner classes in the backward hooks #7062
+68
−68
Conversation
This file contains 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 part of the effort to improve Deepspeed performance when using PyTorch compile. There is a known bug in torch.compile (pytorch/pytorch#128942) which causes a graph break when a inner class is defined within a method that is being compiled and logs the following message: [__graph_breaks] torch._dynamo.exc.Unsupported: missing: LOAD_BUILD_CLASS This is the case with the inner classes PreBackwardFunctionForModule and PostBackwardFunctionModule. While there is an open PR for this (pytorch/pytorch#133805), we can solve the issue by moving the inner classes into the initialization code. No graph breaks and the corresponding logs are produced anymore.
tjruwase
approved these changes
Feb 23, 2025
|
@deepcharm, thanks for the PR. Please resolve the DCO |
|
@deepcharm, gentle ping on DCO blocker |
Propagate API change. Signed-off-by: Olatunji Ruwase <olruwase@microsoft.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
- add zero2 test - minor fix with transformer version update & ds master merge. Signed-off-by: inkcherry <mingzhi.liu@intel.com> Co-authored-by: Olatunji Ruwase <olruwase@microsoft.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
bf16 with moe refresh optimizer state from bf16 ckpt will raise IndexError: list index out of range Signed-off-by: shaomin <wukon1992@gmail.com> Co-authored-by: shaomin <wukon1992@gmail.com> Co-authored-by: Hongwei Chen <33092912+hwchen2017@users.noreply.github.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
**Auto-generated PR to update version.txt after a DeepSpeed release** Released version - 0.16.4 Author - @loadams Co-authored-by: loadams <loadams@users.noreply.github.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
@jeffra and I fixed this many years ago, so bringing this doc to a correct state. --------- Signed-off-by: Stas Bekman <stas@stason.org> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
Description This PR includes Tecorigin SDAA accelerator support. With this PR, DeepSpeed supports SDAA as backend for training tasks. --------- Signed-off-by: siqi <siqi@tecorigin.com> Co-authored-by: siqi <siqi@tecorigin.com> Co-authored-by: Olatunji Ruwase <olruwase@microsoft.com> Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
More information on libuv in pytorch: https://pytorch.org/tutorials/intermediate/TCPStore_libuv_backend.html Issue tracking the prevalence of the error on Windows (unresolved at the time of this PR): pytorch/pytorch#139990 LibUV github: https://github.com/libuv/libuv Windows error: ``` File "C:\hostedtoolcache\windows\Python\3.12.7\x64\Lib\site-packages\torch\distributed\rendezvous.py", line 189, in _create_c10d_store return TCPStore( ^^^^^^^^^ RuntimeError: use_libuv was requested but PyTorch was build without libuv support ``` use_libuv isn't well supported on Windows in pytorch <2.4, so we need to guard around this case. --------- Signed-off-by: Logan Adams <loadams@microsoft.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
Signed-off-by: Logan Adams <loadams@microsoft.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
Signed-off-by: Logan Adams <loadams@microsoft.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
Following changes in Pytorch trace rules , my previous PR to avoid graph breaks caused by logger is no longer relevant. So instead I've added this functionality to torch dynamo - pytorch/pytorch@16ea0dd This commit allows the user to config torch to ignore logger methods and avoid associated graph breaks. To enable ignore logger methods - os.environ["DISABLE_LOGS_WHILE_COMPILING"] = "1" To ignore logger methods except for a specific method / methods (for example, info and isEnabledFor) - os.environ["DISABLE_LOGS_WHILE_COMPILING"] = "1" and os.environ["LOGGER_METHODS_TO_EXCLUDE_FROM_DISABLE"] = "info, isEnabledFor" Signed-off-by: ShellyNR <shelly.nahir@live.biu.ac.il> Co-authored-by: snahir <snahir@habana.ai> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
The partition tensor doesn't need to move to the current device when meta load is used. Signed-off-by: Lai, Yejing <yejing.lai@intel.com> Co-authored-by: Olatunji Ruwase <olruwase@microsoft.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
…t` (deepspeedai#7069) With future changes coming to pip/python/etc, we need to modify to no longer call `python setup.py ...` and replace it instead: https://packaging.python.org/en/latest/guides/modernize-setup-py-project/#should-setup-py-be-deleted  This means we need to install the build package which is added here as well. Additionally, we pass the `--sdist` flag to only build the sdist rather than the wheel as well here. --------- Signed-off-by: Logan Adams <loadams@microsoft.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
…eepspeedai#7076) This reverts commit 8577bd2. Fixes: deepspeedai#7072 Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
deepcharm
force-pushed
the
avoid-graph-break-caused-by-inner-classes
branch
from
561bd6e to
d4efb6a
Compare
deepcharm
requested review from
jomayeri,
hwchen2017 and
loadams
as code owners
@tjruwase DCO should be fine now, sorry for the delay |
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.
This PR is part of the effort to improve Deepspeed performance when using PyTorch compile.
There is a known bug in torch.compile which causes a graph break when an inner class is defined within
a method that is being compiled. The following would then appear in the log:
[__graph_breaks] torch._dynamo.exc.Unsupported: missing: LOAD_BUILD_CLASSThis is the case with the inner classes
PreBackwardFunctionForModuleandPostBackwardFunctionModule.While there is an open PyTorch PR#133805 for this, we can solve the issue by moving the inner classes into the initialization code.
No graph breaks and the corresponding logs are produced anymore.