Skip to content

[Accuracy diff No.40-41] [BIT] fix cosine_similarity accuracy error when x1, x2 need to broadcast #73014

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 8 commits into from
Jun 6, 2025
Merged

[Accuracy diff No.40-41] [BIT] fix cosine_similarity accuracy error when x1, x2 need to broadcast #73014

merged 8 commits into from
Jun 6, 2025

Conversation

Cutelemon6
Copy link
Contributor

@Cutelemon6 Cutelemon6 commented May 29, 2025 •
edited
Loading

PR Category

Execute Infrastructure

PR Types

Bug fixes

Description

Fix bug when x1.shape != x2.shape, resultig wrong w1 or w2 which does sumation over the axis that need to broadcast.
A minimum reproducible case compared to torch

a = numpy.array([[1, 2]])
b = numpy.array([[2], [4]])

pa = paddle.to_tensor(a, dtype=paddle.float64)
pb = paddle.to_tensor(b, dtype=paddle.float64)
ta = torch.tensor(a, dtype=torch.float64)
tb = torch.tensor(b, dtype=torch.float64)

pr = paddle.nn.functional.cosine_similarity(pa, pb)
tr = torch.nn.functional.cosine_similarity(ta, tb)
print("paddle: ", pr)
print("torch: ", tr)

output

paddle:  Tensor(shape=[2], dtype=float64, place=Place(gpu:0), stop_gradient=True,
       [1.34164079, 1.34164079])
torch:  tensor([0.9487, 0.9487], dtype=torch.float64)

in this case, paddle implementation

w12 = sum(paddle.multiply(x1, x2), axis=axis)
w1 = sum(paddle.multiply(x1, x1), axis=axis)
w2 = sum(paddle.multiply(x2, x2), axis=axis)
n12 = sqrt(clip(w1 * w2, min=eps * eps))
cos_sim = w12 / n12
return cos_sim

w2 is [4, 16], but should be [8, 32] after broadcast and sumation.

All cases (24 cases in total) have passed both in gpu and cpu environments.

gpu:
cosine_sim-gpu

cpu:
cosine_sim-cpu

@paddle-bot Paddle-bot
Copy link

paddle-bot bot commented May 29, 2025

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

@paddle-bot paddle-bot bot added the contributor External developers label May 29, 2025
@Cutelemon6 Cutelemon6 changed the title [Accuracy diff No.40-41] fix cosine_similarity accuracy error when x1, x2 need to broadcast [Accuracy diff No.40-41] [BIT] fix cosine_similarity accuracy error when x1, x2 need to broadcast May 29, 2025
@luotao1 luotao1 mentioned this pull request May 30, 2025
Open
@Cutelemon6 Cutelemon6 mentioned this pull request May 30, 2025
Merged
@codecov-commenter
Copy link

codecov-commenter commented May 30, 2025 •
edited
Loading

Codecov Report

All modified and coverable lines are covered by tests ✅

Please upload report for BASE (develop@e9ad101). Learn more about missing BASE report.

Additional details and impacted files 
@@             Coverage Diff             @@
##             develop    #73014   +/-   ##
===========================================
  Coverage           ?   100.00%           
===========================================
  Files              ?         1           
  Lines              ?         6           
  Branches           ?         0           
===========================================
  Hits               ?         6           
  Misses             ?         0           
  Partials           ?         0

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@lshpku lshpku merged commit 79f3f5f into PaddlePaddle:develop Jun 6, 2025
49 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@lshpku lshpku lshpku approved these changes

Assignees
No one assigned
Labels
contributor External developers
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@Cutelemon6 @codecov-commenter @lshpku