fix(jax-backend): fix failing test for sum #27087

Aaryan562 · 2023-10-21T05:01:51Z

fix failing test for sum

github-actions

PR Compliance Checks

Thank you for your Pull Request! We have run several checks on this pull request in order to make sure it's suitable for merging into this project. The results are listed in the following section.

Issue Reference

In order to be considered for merging, the pull request description must refer to a specific issue number. This is described in our contributing guide and our PR template.
This check is looking for a phrase similar to: "Fixes #XYZ" or "Resolves #XYZ" where XYZ is the issue number that this PR is meant to address.

github-actions · 2023-10-21T05:51:08Z

Thank you for this PR, here is the CI results:

This pull request does not result in any additional test failures. Congratulations!

AnnaTz · 2023-10-23T16:43:13Z

Hi @Aaryan562, I'm not sure we should declare complex as unsupported for the jax backend in general, because the error you are trying to fix only occurs when test_gradients=True.

@vedpatwardhan, what would you suggest in this case?
Fot the record, I ran the test and the full error is this:

ivy.utils.exceptions.IvyBackendException: jax: execute_with_gradients: grad requires real-valued outputs (output dtype that is a sub-dtype of np.floating), but got complex128. For holomorphic differentiation, pass holomorphic=True. For differentiation of non-holomorphic functions involving complex outputs, use jax.vjp directly.

vedpatwardhan · 2023-10-25T04:06:32Z

@vedpatwardhan, what would you suggest in this case? Fot the record, I ran the test and the full error is this:

ivy.utils.exceptions.IvyBackendException: jax: execute_with_gradients: grad requires real-valued outputs (output dtype that is a sub-dtype of np.floating), but got complex128. For holomorphic differentiation, pass holomorphic=True. For differentiation of non-holomorphic functions involving complex outputs, use jax.vjp directly.

Ideally we shouldn't be running gradient tests when the input is a complex dtype. We also have a check to avoid that from happening, could you please look into why the check isn't working for this test @Aaryan562? Thanks 😄

Aaryan562 · 2023-10-26T07:58:59Z

@vedpatwardhan, what would you suggest in this case? Fot the record, I ran the test and the full error is this:
ivy.utils.exceptions.IvyBackendException: jax: execute_with_gradients: grad requires real-valued outputs (output dtype that is a sub-dtype of np.floating), but got complex128. For holomorphic differentiation, pass holomorphic=True. For differentiation of non-holomorphic functions involving complex outputs, use jax.vjp directly.
Ideally we shouldn't be running gradient tests when the input is a complex dtype. We also have a check to avoid that from happening, could you please look into why the check isn't working for this test @Aaryan562? Thanks 😄

@vedpatwardhan @AnnaTz The paddle version has been changed to 2.5.2, which might create error in the paddle backend. The jax backend for the complex datatypes has been resolved. Kindly check!

ivy_tests/test_ivy/test_functional/test_core/test_statistical.py

vedpatwardhan

Hey @Aaryan562, just requested a final change, thanks 😄

vedpatwardhan · 2023-10-30T11:16:14Z

ivy_tests/test_ivy/test_functional/test_core/test_statistical.py

@@ -342,6 +342,9 @@ def test_sum(
    if "torch" in backend_fw:
        assume(not test_flags.as_variable[0])
        assume(not test_flags.test_gradients)
+    if "jax" in backend_fw and castable_dtype in ["complex32", "complex64"]:


I think this should be "complex32", "complex128" instead, what do you think?

I think this should be "complex32", "complex128" instead, what do you think?

Ivy support complex64 and complex128 so maybe we should cater to that?

…ackend

ivy/functional/backends/paddle/statistical.py

Aaryan562 · 2023-10-30T19:47:51Z

ivy/functional/backends/paddle/statistical.py

@@ -167,7 +166,10 @@ def std(
    return _std(x, axis, correction, keepdims).cast(x.dtype)


-@with_unsupported_dtypes({"2.5.2 and below": ("int8", "uint8")}, backend_version)
+@with_supported_dtypes(


Also the paddle backend gave this error so changed its decorator ivy.utils.exceptions.IvyBackendException: paddle: execute_with_gradients: paddle: nested_map: paddle: mean: (NotFound) The kernel with key (CPU, Undefined(AnyLayout), int32) of kernel mean is not registered. Selected wrong DataType int32. Paddle support following DataTypes: float64, complex128, float32, complex64, bool
I dont know why mean is getting triggered even when we are dealing with sum function.

mean is actually applied for the gradient tests here, so both mean and sum are called when running the gradient tests for sum. We should probably we fixing mean first if there's an issue with mean. Thanks @Aaryan562 😄
Also, does paddle not natively support int dtypes for its mean and sum functions (cc @MahmoudAshraf97)?

[int8, int16, uint8] are not supported for mean and [int8, uint8] for sum

oh okay, ig then int32 and int64 can also be marked as supported @Aaryan562 😄
(probably specifying the unsupported dtypes is more convenient that specifying the supported dtypes for this function)

ivy.utils.exceptions.IvyBackendException: paddle: execute_with_gradients: paddle: nested_map: paddle: mean: (NotFound) The kernel with key (CPU, Undefined(AnyLayout), int32) of kernel mean is not registered. Selected wrong DataType int32. Paddle support following DataTypes: float64, complex128, float32, complex64, bool

@vedpatwardhan But then when supporting int32 and int64 i am getting the above error.

I further debugged I got to know it was when my test_gradients=True. So I changed my test_sum to make it test_gradients=False and all the test cases passed succesfully. But what could be the reason for it?

vedpatwardhan · 2023-10-31T04:17:46Z

ivy/functional/backends/paddle/statistical.py

@@ -167,7 +166,10 @@ def std(
    return _std(x, axis, correction, keepdims).cast(x.dtype)


-@with_unsupported_dtypes({"2.5.2 and below": ("int8", "uint8")}, backend_version)
+@with_supported_dtypes(


mean is actually applied for the gradient tests here, so both mean and sum are called when running the gradient tests for sum. We should probably we fixing mean first if there's an issue with mean. Thanks @Aaryan562 😄
Also, does paddle not natively support int dtypes for its mean and sum functions (cc @MahmoudAshraf97)?

vedpatwardhan

lgtm! Feel free to merge, thanks @Aaryan562 😄

added decorators to sum

78ef35a

Aaryan562 requested a review from AnnaTz October 21, 2023 05:01

github-actions bot reviewed Oct 21, 2023

View reviewed changes

ivy-seed assigned abdulasiraj Oct 21, 2023

Aaryan562 added 3 commits October 25, 2023 21:06

Merge branch 'unifyai:main' into sum_stat

e538883

Merge branch 'unifyai:main' into sum_stat

8a30fe4

reverted the changes of adding decorato to jax backend

4b7b9b4

vedpatwardhan reviewed Oct 27, 2023

View reviewed changes

ivy_tests/test_ivy/test_functional/test_core/test_statistical.py Outdated Show resolved Hide resolved

minor change to focus on jax and complex dtypes

8ee2fcf

Aaryan562 requested a review from vedpatwardhan October 27, 2023 07:12

vedpatwardhan suggested changes Oct 30, 2023

View reviewed changes

Aaryan562 added 2 commits October 31, 2023 00:38

Merge branch 'unifyai:main' into sum_stat

df9aa03

solved error in datatype of paddle and changes complex dtype in jax b…

bcc31c8

…ackend

Aaryan562 requested a review from vedpatwardhan October 30, 2023 19:37

Aaryan562 commented Oct 30, 2023

View reviewed changes

ivy/functional/backends/paddle/statistical.py Outdated Show resolved Hide resolved

Aaryan562 commented Oct 30, 2023

View reviewed changes

ivy-leaves added the PaddlePaddle Backend Developing the Paddle Paddle Backend. label Oct 30, 2023

vedpatwardhan suggested changes Oct 31, 2023

View reviewed changes

made test_gradients to false

9b0d2db

Aaryan562 requested a review from vedpatwardhan November 2, 2023 08:25

vedpatwardhan approved these changes Nov 3, 2023

View reviewed changes

Aaryan562 merged commit 7af38de into ivy-llc:main Nov 3, 2023
176 of 273 checks passed

Aaryan562 deleted the sum_stat branch November 11, 2023 09:29

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(jax-backend): fix failing test for sum #27087

fix(jax-backend): fix failing test for sum #27087

Aaryan562 commented Oct 21, 2023

github-actions bot left a comment

github-actions bot commented Oct 21, 2023

AnnaTz commented Oct 23, 2023 •

edited

Loading

vedpatwardhan commented Oct 25, 2023

Aaryan562 commented Oct 26, 2023

vedpatwardhan left a comment

vedpatwardhan Oct 30, 2023

Aaryan562 Oct 30, 2023 •

edited

Loading

Aaryan562 Oct 30, 2023

vedpatwardhan Oct 31, 2023

MahmoudAshraf97 Oct 31, 2023

vedpatwardhan Nov 1, 2023

Aaryan562 Nov 1, 2023 •

edited

Loading

Aaryan562 Nov 1, 2023 •

edited

Loading

vedpatwardhan Oct 31, 2023

vedpatwardhan left a comment

fix(jax-backend): fix failing test for sum #27087

fix(jax-backend): fix failing test for sum #27087

Conversation

Aaryan562 commented Oct 21, 2023

github-actions bot left a comment

Choose a reason for hiding this comment

PR Compliance Checks

Issue Reference

github-actions bot commented Oct 21, 2023

AnnaTz commented Oct 23, 2023 • edited Loading

vedpatwardhan commented Oct 25, 2023

Aaryan562 commented Oct 26, 2023

vedpatwardhan left a comment

Choose a reason for hiding this comment

vedpatwardhan Oct 30, 2023

Choose a reason for hiding this comment

Aaryan562 Oct 30, 2023 • edited Loading

Choose a reason for hiding this comment

Aaryan562 Oct 30, 2023

Choose a reason for hiding this comment

vedpatwardhan Oct 31, 2023

Choose a reason for hiding this comment

MahmoudAshraf97 Oct 31, 2023

Choose a reason for hiding this comment

vedpatwardhan Nov 1, 2023

Choose a reason for hiding this comment

Aaryan562 Nov 1, 2023 • edited Loading

Choose a reason for hiding this comment

Aaryan562 Nov 1, 2023 • edited Loading

Choose a reason for hiding this comment

vedpatwardhan Oct 31, 2023

Choose a reason for hiding this comment

vedpatwardhan left a comment

Choose a reason for hiding this comment

AnnaTz commented Oct 23, 2023 •

edited

Loading

Aaryan562 Oct 30, 2023 •

edited

Loading

Aaryan562 Nov 1, 2023 •

edited

Loading

Aaryan562 Nov 1, 2023 •

edited

Loading