[Pytorch][Vulkan] layer_norm #112322

copyrightly · 2023-10-28T18:08:10Z

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.

(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

pytorch-bot · 2023-10-28T18:08:14Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/112322

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit e2b1f14 with merge base eb8af4d ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

facebook-github-bot · 2023-10-28T18:08:22Z

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 · 2023-10-28T19:32:31Z

This pull request was exported from Phabricator. Differential Revision: D50436726

facebook-github-bot · 2023-10-28T19:32:49Z

This pull request was exported from Phabricator. Differential Revision: D50436726

Skylion007 · 2023-10-29T17:48:42Z

aten/src/ATen/native/vulkan/ops/Layernorm.cpp

-
-  return convert(v_output);
+
+  std::vector<int64_t> dims_to_reduce;


Suggested change

std::vector<int64_t> dims_to_reduce;

std::vector<int64_t> dims_to_reduce;

dims_to_reduce.reserve(normalized_shape.size());

Thank you @Skylion007! I will fix it in a subsequent PR.

facebook-github-bot · 2023-10-30T19:17:50Z

@pytorchbot merge

(Initiating merge automatically since Phabricator Diff has merged)

pytorchmergebot · 2023-10-30T19:21:09Z

Merge started

Your 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

Advanced Debugging

Check the merge workflow status
here

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

pytorch-bot bot added ciflow/periodic Trigger jobs ran periodically on master (periodic.yml) on the PR module: vulkan release notes: vulkan release notes category labels Oct 28, 2023

facebook-github-bot added the fb-exported label Oct 28, 2023

facebook-github-bot force-pushed the export-D50436726 branch from 52257d0 to cffba57 Compare October 28, 2023 19:32

copyrightly force-pushed the export-D50436726 branch from cffba57 to e2b1f14 Compare October 28, 2023 19:32

Skylion007 reviewed Oct 29, 2023

View reviewed changes

yipjustin approved these changes Oct 30, 2023

View reviewed changes

pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label Oct 30, 2023

pytorchmergebot added the merging label Oct 30, 2023

pytorchmergebot added Merged and removed merging labels Oct 30, 2023

pytorchmergebot closed this in 503955f Oct 30, 2023

facebook-github-bot deleted the export-D50436726 branch November 3, 2023 14:26

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Pytorch][Vulkan] layer_norm #112322

[Pytorch][Vulkan] layer_norm #112322

Uh oh!

copyrightly commented Oct 28, 2023

Uh oh!

pytorch-bot bot commented Oct 28, 2023 •

edited

Loading

Uh oh!

facebook-github-bot commented Oct 28, 2023

Uh oh!

facebook-github-bot commented Oct 28, 2023

Uh oh!

facebook-github-bot commented Oct 28, 2023

Uh oh!

Skylion007 Oct 29, 2023

Uh oh!

copyrightly Oct 30, 2023

Uh oh!

facebook-github-bot commented Oct 30, 2023

Uh oh!

pytorchmergebot commented Oct 30, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants


		return convert(v_output);

		std::vector<int64_t> dims_to_reduce;

	std::vector<int64_t> dims_to_reduce;
	std::vector<int64_t> dims_to_reduce;
	dims_to_reduce.reserve(normalized_shape.size());

[Pytorch][Vulkan] layer_norm #112322

[Pytorch][Vulkan] layer_norm #112322

Uh oh!

Conversation

copyrightly commented Oct 28, 2023

Uh oh!

pytorch-bot bot commented Oct 28, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/112322

✅ No Failures

Uh oh!

facebook-github-bot commented Oct 28, 2023

Uh oh!

facebook-github-bot commented Oct 28, 2023

Uh oh!

facebook-github-bot commented Oct 28, 2023

Uh oh!

Skylion007 Oct 29, 2023

Choose a reason for hiding this comment

Uh oh!

copyrightly Oct 30, 2023

Choose a reason for hiding this comment

Uh oh!

facebook-github-bot commented Oct 30, 2023

Uh oh!

pytorchmergebot commented Oct 30, 2023

Merge started

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

pytorch-bot bot commented Oct 28, 2023 •

edited

Loading