[SyncBatchNorm] Support running with low precision parameters #98332
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
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/98332
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ✅ No FailuresAs of commit 285e049: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Apr 4, 2023
awgu
added a commit
that referenced
this pull request
Apr 4, 2023
ghstack-source-id: 3fc48950d1ebc41bb25bac9939a1c1e3e315d39d Pull Request resolved: #98332
…ded" This PR fixes #96203. **Details** When using `nn.SyncBatchNorm` with the model converted to FP16, there is a dtype discrepancy in the `SyncBatchNorm.forward()` causing an error like: ``` File "/.../pytorch/torch/nn/modules/_functions.py", line 91, in forward mean, invstd = torch.batch_norm_gather_stats_with_counts( RuntimeError: Expected counts to have type Half but got Float ``` [`torch.batch_norm_gather_stats_with_counts()`](https://github.com/pytorch/pytorch/blob/fe9da29842a07a1f44d6b8c2a4c75053da9e84d0/torch/nn/modules/_functions.py#L88-L97) requires the `running_mean`, `running_var`, and `counts` to have the same dtype. However, when the model has been converted to FP16, only `running_mean` and `running_var` use FP16, while the `counts` are in FP32 due to [`mean` being in FP32](https://github.com/pytorch/pytorch/blob/fe9da29842a07a1f44d6b8c2a4c75053da9e84d0/torch/nn/modules/_functions.py#L25-L30). This PR resolves this by casting `counts` from FP32 to FP16 instead of the alternative to cast `mean` and `invstd` from FP32 to FP16. Moreover, for the backward, this PR casts `weight` from FP16 to FP32 to match the dtype of `mean` and `invstd` as required by `torch.batch_norm_backward_elemt()` instead of the alternative to cast `mean` and `invstd` from FP32 to FP16. **Test Plan** I dug up this run command from 2021: ``` WORLD_SIZE=2 BACKEND=nccl python -m pytest test/distributed/test_distributed_spawn.py -k test_DistributedDataParallel_SyncBatchNorm_half -vs ``` [ghstack-poisoned]
awgu
added a commit
that referenced
this pull request
Apr 4, 2023
ghstack-source-id: e8a775d674c7d1c0f17b9b2348c9ca6184a59d54 Pull Request resolved: #98332
awgu
requested review from
mrshenli,
zhaojuanmao,
H-Huang,
kwen2501,
wanchaol,
fegin,
albanD,
jbschlosser and
mikaylagawarecki
as code owners
awgu
added
topic: improvements
awgu
changed the title
running_mean.dtype if needed…ers" This PR fixes #96203. **Details** When using `nn.SyncBatchNorm` with the model converted to FP16, there is a dtype discrepancy in the `SyncBatchNorm.forward()` causing an error like: ``` File "/.../pytorch/torch/nn/modules/_functions.py", line 91, in forward mean, invstd = torch.batch_norm_gather_stats_with_counts( RuntimeError: Expected counts to have type Half but got Float ``` [`torch.batch_norm_gather_stats_with_counts()`](https://github.com/pytorch/pytorch/blob/fe9da29842a07a1f44d6b8c2a4c75053da9e84d0/torch/nn/modules/_functions.py#L88-L97) requires the `running_mean`, `running_var`, and `counts` to have the same dtype. However, when the model has been converted to FP16, only `running_mean` and `running_var` use FP16, while the `counts` are in FP32 due to [`mean` being in FP32](https://github.com/pytorch/pytorch/blob/fe9da29842a07a1f44d6b8c2a4c75053da9e84d0/torch/nn/modules/_functions.py#L25-L30). This PR resolves this by casting `counts` from FP32 to FP16 instead of the alternative to cast `mean` and `invstd` from FP32 to FP16. Moreover, for the backward, this PR casts `weight` from FP16 to FP32 to match the dtype of `mean` and `invstd` as required by `torch.batch_norm_backward_elemt()` instead of the alternative to cast `mean` and `invstd` from FP32 to FP16. **Test Plan** I dug up this run commands from 2021: For `world_size` in `{1,2}` and `backend` in `{nccl, gloo}`: ``` WORLD_SIZE=world_size BACKEND=backend python -m pytest test/distributed/test_distributed_spawn.py -k test_DistributedDataParallel_SyncBatchNorm_half -vs ``` [ghstack-poisoned]
awgu
added a commit
that referenced
this pull request
Apr 4, 2023
ghstack-source-id: a804c303e0ebe15f4fb0680737c25992d2c44da8 Pull Request resolved: #98332
|
@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 |
|
@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 |
ZainRizvi
pushed a commit
that referenced
this pull request
Apr 19, 2023
This PR fixes #96203. **Details** When using `nn.SyncBatchNorm` with the model converted to FP16, there is a dtype discrepancy in the `SyncBatchNorm.forward()` causing an error like: ``` File "/.../pytorch/torch/nn/modules/_functions.py", line 91, in forward mean, invstd = torch.batch_norm_gather_stats_with_counts( RuntimeError: Expected counts to have type Half but got Float ``` [`torch.batch_norm_gather_stats_with_counts()`](https://github.com/pytorch/pytorch/blob/fe9da29842a07a1f44d6b8c2a4c75053da9e84d0/torch/nn/modules/_functions.py#L88-L97) requires the `running_mean`, `running_var`, and `counts` to have the same dtype. However, when the model has been converted to FP16, only `running_mean` and `running_var` use FP16, while the `counts` are in FP32 due to [`mean` being in FP32](https://github.com/pytorch/pytorch/blob/fe9da29842a07a1f44d6b8c2a4c75053da9e84d0/torch/nn/modules/_functions.py#L25-L30). This PR resolves this by casting `counts` from FP32 to FP16 instead of the alternative to cast `mean` and `invstd` from FP32 to FP16. Moreover, for the backward, this PR casts `weight` from FP16 to FP32 to match the dtype of `mean` and `invstd` as required by `torch.batch_norm_backward_elemt()` instead of the alternative to cast `mean` and `invstd` from FP32 to FP16. **Test Plan** I dug up this run command from 2021: For `world_size` in `{1,2}` and `backend` in `{nccl, gloo}`: ``` WORLD_SIZE=world_size BACKEND=backend python -m pytest test/distributed/test_distributed_spawn.py -k test_DistributedDataParallel_SyncBatchNorm_half -vs ``` Pull Request resolved: #98332 Approved by: https://github.com/rohan-varma
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):
requires_grad_mask#98299_use_sharded_views()forSHARD_GRAD_OP#98250requires_gradforuse_orig_params=True#98221This PR fixes #96203.
Details
When using
nn.SyncBatchNormwith the model converted to FP16, there is a dtype discrepancy in theSyncBatchNorm.forward()causing an error like:torch.batch_norm_gather_stats_with_counts()requires therunning_mean,running_var, andcountsto have the same dtype. However, when the model has been converted to FP16, onlyrunning_meanandrunning_varuse FP16, while thecountsare in FP32 due tomeanbeing in FP32. This PR resolves this by castingcountsfrom FP32 to FP16 instead of the alternative to castmeanandinvstdfrom FP32 to FP16.Moreover, for the backward, this PR casts
weightfrom FP16 to FP32 to match the dtype ofmeanandinvstdas required bytorch.batch_norm_backward_elemt()instead of the alternative to castmeanandinvstdfrom FP32 to FP16.Test Plan
I dug up this run command from 2021:
For
world_sizein{1,2}andbackendin{nccl, gloo}: