[Pytorch][Vulkan] layer_norm #112322
Closed
[Pytorch][Vulkan] layer_norm #112322
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/112322
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit e2b1f14 with merge base eb8af4d ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
ciflow/periodic
|
This pull request was exported from Phabricator. Differential Revision: D50436726 |
Summary: Generalize [layer_norm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html) to all tensors of 2d to 4d. Using the mean and var operators in this diff stack, we can compute the layer_norm directly and remove the old shader file `layernorm.glsl`. ``` (input - input.mean(normalized_shape, keepdim=True)) / torch.sqrt(input.var(normalized_shape, correction=0, keepdims = True) + eps) * weight + bias ``` Test Plan: ``` [luwei@devbig984.prn1 /data/users/luwei/fbsource (0a5028d8c)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*layer_norm*" Building: finished in 0.1 sec (100%) 339/339 jobs, 0/339 updated Total time: 0.1 sec BUILD SUCCEEDED Running main() from third-party/googletest/1.11.0/googletest/googletest/src/gtest_main.cc Note: Google Test filter = *layer_norm* [==========] Running 4 tests from 1 test suite. [----------] Global test environment set-up. [----------] 4 tests from VulkanAPITest [ RUN ] VulkanAPITest.layer_norm_invalid_inputs [ OK ] VulkanAPITest.layer_norm_invalid_inputs (69 ms) [ RUN ] VulkanAPITest.layer_norm_2d [ OK ] VulkanAPITest.layer_norm_2d (288 ms) [ RUN ] VulkanAPITest.layer_norm_3d [ OK ] VulkanAPITest.layer_norm_3d (302 ms) [ RUN ] VulkanAPITest.layer_norm_4d [ OK ] VulkanAPITest.layer_norm_4d (8 ms) [----------] 4 tests from VulkanAPITest (668 ms total) [----------] Global test environment tear-down [==========] 4 tests from 1 test suite ran. (668 ms total) [ PASSED ] 4 tests. ``` Reviewed By: yipjustin Differential Revision: D50436726
facebook-github-bot
force-pushed
the
export-D50436726
branch
from
52257d0
to
cffba57
Compare
facebook-github-bot
pushed a commit
that referenced
this pull request
Oct 28, 2023
Summary: Generalize [layer_norm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html) to all tensors of 2d to 4d. Using the mean and var operators in this diff stack, we can compute the layer_norm directly and remove the old shader file `layernorm.glsl`. ``` (input - input.mean(normalized_shape, keepdim=True)) / torch.sqrt(input.var(normalized_shape, correction=0, keepdims = True) + eps) * weight + bias ``` Test Plan: ``` [luwei@devbig984.prn1 /data/users/luwei/fbsource (0a5028d8c)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*layer_norm*" Building: finished in 0.1 sec (100%) 339/339 jobs, 0/339 updated Total time: 0.1 sec BUILD SUCCEEDED Running main() from third-party/googletest/1.11.0/googletest/googletest/src/gtest_main.cc Note: Google Test filter = *layer_norm* [==========] Running 4 tests from 1 test suite. [----------] Global test environment set-up. [----------] 4 tests from VulkanAPITest [ RUN ] VulkanAPITest.layer_norm_invalid_inputs [ OK ] VulkanAPITest.layer_norm_invalid_inputs (69 ms) [ RUN ] VulkanAPITest.layer_norm_2d [ OK ] VulkanAPITest.layer_norm_2d (288 ms) [ RUN ] VulkanAPITest.layer_norm_3d [ OK ] VulkanAPITest.layer_norm_3d (302 ms) [ RUN ] VulkanAPITest.layer_norm_4d [ OK ] VulkanAPITest.layer_norm_4d (8 ms) [----------] 4 tests from VulkanAPITest (668 ms total) [----------] Global test environment tear-down [==========] 4 tests from 1 test suite ran. (668 ms total) [ PASSED ] 4 tests. ``` Reviewed By: yipjustin Differential Revision: D50436726
|
This pull request was exported from Phabricator. Differential Revision: D50436726 |
copyrightly
force-pushed
the
export-D50436726
branch
from
cffba57
to
e2b1f14
Compare
|
This pull request was exported from Phabricator. Differential Revision: D50436726 |
Skylion007
reviewed
Oct 29, 2023
|
|
||
| return convert(v_output); | ||
|
|
||
| std::vector<int64_t> dims_to_reduce; |
There was a problem hiding this comment.
Suggested change
| std::vector<int64_t> dims_to_reduce; | |
| std::vector<int64_t> dims_to_reduce; | |
| dims_to_reduce.reserve(normalized_shape.size()); |
There was a problem hiding this comment.
Thank you @Skylion007! I will fix it in a subsequent PR.
yipjustin
approved these changes
Oct 30, 2023
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
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 |
xuhancn
pushed a commit
to xuhancn/pytorch
that referenced
this pull request
Nov 7, 2023
Summary: Generalize [layer_norm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html) to all tensors of 2d to 4d. Using the mean and var operators in this diff stack, we can compute the layer_norm directly and remove the old shader file `layernorm.glsl`. ``` (input - input.mean(normalized_shape, keepdim=True)) / torch.sqrt(input.var(normalized_shape, correction=0, keepdims = True) + eps) * weight + bias ``` Test Plan: ``` [luwei@devbig984.prn1 /data/users/luwei/fbsource (0a5028d8c)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*layer_norm*" Building: finished in 0.1 sec (100%) 339/339 jobs, 0/339 updated Total time: 0.1 sec BUILD SUCCEEDED Running main() from third-party/googletest/1.11.0/googletest/googletest/src/gtest_main.cc Note: Google Test filter = *layer_norm* [==========] Running 4 tests from 1 test suite. [----------] Global test environment set-up. [----------] 4 tests from VulkanAPITest [ RUN ] VulkanAPITest.layer_norm_invalid_inputs [ OK ] VulkanAPITest.layer_norm_invalid_inputs (69 ms) [ RUN ] VulkanAPITest.layer_norm_2d [ OK ] VulkanAPITest.layer_norm_2d (288 ms) [ RUN ] VulkanAPITest.layer_norm_3d [ OK ] VulkanAPITest.layer_norm_3d (302 ms) [ RUN ] VulkanAPITest.layer_norm_4d [ OK ] VulkanAPITest.layer_norm_4d (8 ms) [----------] 4 tests from VulkanAPITest (668 ms total) [----------] Global test environment tear-down [==========] 4 tests from 1 test suite ran. (668 ms total) [ PASSED ] 4 tests. ``` Reviewed By: yipjustin Differential Revision: D50436726 Pull Request resolved: pytorch#112322 Approved by: https://github.com/yipjustin
Skylion007
pushed a commit
to Skylion007/pytorch
that referenced
this pull request
Nov 14, 2023
Summary: Generalize [layer_norm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html) to all tensors of 2d to 4d. Using the mean and var operators in this diff stack, we can compute the layer_norm directly and remove the old shader file `layernorm.glsl`. ``` (input - input.mean(normalized_shape, keepdim=True)) / torch.sqrt(input.var(normalized_shape, correction=0, keepdims = True) + eps) * weight + bias ``` Test Plan: ``` [luwei@devbig984.prn1 /data/users/luwei/fbsource (0a5028d8c)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*layer_norm*" Building: finished in 0.1 sec (100%) 339/339 jobs, 0/339 updated Total time: 0.1 sec BUILD SUCCEEDED Running main() from third-party/googletest/1.11.0/googletest/googletest/src/gtest_main.cc Note: Google Test filter = *layer_norm* [==========] Running 4 tests from 1 test suite. [----------] Global test environment set-up. [----------] 4 tests from VulkanAPITest [ RUN ] VulkanAPITest.layer_norm_invalid_inputs [ OK ] VulkanAPITest.layer_norm_invalid_inputs (69 ms) [ RUN ] VulkanAPITest.layer_norm_2d [ OK ] VulkanAPITest.layer_norm_2d (288 ms) [ RUN ] VulkanAPITest.layer_norm_3d [ OK ] VulkanAPITest.layer_norm_3d (302 ms) [ RUN ] VulkanAPITest.layer_norm_4d [ OK ] VulkanAPITest.layer_norm_4d (8 ms) [----------] 4 tests from VulkanAPITest (668 ms total) [----------] Global test environment tear-down [==========] 4 tests from 1 test suite ran. (668 ms total) [ PASSED ] 4 tests. ``` Reviewed By: yipjustin Differential Revision: D50436726 Pull Request resolved: pytorch#112322 Approved by: https://github.com/yipjustin
andreigh
pushed a commit
to andreigh/pytorch
that referenced
this pull request
Nov 19, 2023
Summary: Generalize [layer_norm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html) to all tensors of 2d to 4d. Using the mean and var operators in this diff stack, we can compute the layer_norm directly and remove the old shader file `layernorm.glsl`. ``` (input - input.mean(normalized_shape, keepdim=True)) / torch.sqrt(input.var(normalized_shape, correction=0, keepdims = True) + eps) * weight + bias ``` Test Plan: ``` [luwei@devbig984.prn1 /data/users/luwei/fbsource (0a5028d8c)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*layer_norm*" Building: finished in 0.1 sec (100%) 339/339 jobs, 0/339 updated Total time: 0.1 sec BUILD SUCCEEDED Running main() from third-party/googletest/1.11.0/googletest/googletest/src/gtest_main.cc Note: Google Test filter = *layer_norm* [==========] Running 4 tests from 1 test suite. [----------] Global test environment set-up. [----------] 4 tests from VulkanAPITest [ RUN ] VulkanAPITest.layer_norm_invalid_inputs [ OK ] VulkanAPITest.layer_norm_invalid_inputs (69 ms) [ RUN ] VulkanAPITest.layer_norm_2d [ OK ] VulkanAPITest.layer_norm_2d (288 ms) [ RUN ] VulkanAPITest.layer_norm_3d [ OK ] VulkanAPITest.layer_norm_3d (302 ms) [ RUN ] VulkanAPITest.layer_norm_4d [ OK ] VulkanAPITest.layer_norm_4d (8 ms) [----------] 4 tests from VulkanAPITest (668 ms total) [----------] Global test environment tear-down [==========] 4 tests from 1 test suite ran. (668 ms total) [ PASSED ] 4 tests. ``` Reviewed By: yipjustin Differential Revision: D50436726 Pull Request resolved: pytorch#112322 Approved by: https://github.com/yipjustin
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.
Summary:
Generalize layer_norm to all tensors of 2d to 4d. Using the mean and var operators in this diff stack, we can compute the layer_norm directly and remove the old shader file
layernorm.glsl.Test Plan:
Reviewed By: yipjustin
Differential Revision: D50436726