Fix `backend/mlx/core` and `backend/common/dtypes` for MLX + Improve `integration_tests/numerical_test.py` #19619

Faisal-Alsrheed · 2024-04-26T05:53:59Z

MLX Backend Specific Failures:

Missing Operations: The MLX backend lack implementations for specific operations, including:

mod
rand
rmatmul
ror
rpow
rxor
xor

 KERAS_BACKEND=mlx pytest keras/src/backend/
=================================================================================================================== short test summary info ====================================================================================================================
FAILED keras/src/backend/common/variables_test.py::VariableOperationsTest::test__mod__ - AssertionError: 
FAILED keras/src/backend/common/variables_test.py::VariableOperationsTest::test__rand__ - AttributeError: 'mlx.core.array' object has no attribute '__rand__'. Did you mean: '__and__'?
FAILED keras/src/backend/common/variables_test.py::VariableOperationsTest::test__rmatmul__ - AttributeError: 'mlx.core.array' object has no attribute '__rmatmul__'. Did you mean: '__imatmul__'?
FAILED keras/src/backend/common/variables_test.py::VariableOperationsTest::test__ror__ - AttributeError: 'mlx.core.array' object has no attribute '__ror__'. Did you mean: '__ior__'?
FAILED keras/src/backend/common/variables_test.py::VariableOperationsTest::test__rpow__ - AttributeError: 'mlx.core.array' object has no attribute '__rpow__'. Did you mean: '__ipow__'?
FAILED keras/src/backend/common/variables_test.py::VariableOperationsTest::test__rxor__ - AttributeError: 'mlx.core.array' object has no attribute '__rxor__'. Did you mean: '__ior__'?
FAILED keras/src/backend/common/variables_test.py::VariableOperationsTest::test__xor__ - AttributeError: 'mlx.core.array' object has no attribute '__xor__'. Did you mean: '__ior__'?
========================================================================================================== 7 failed, 413 passed, 91 skipped in 7.92s ===========================================================================================================

codecov-commenter · 2024-04-26T06:00:57Z

Codecov Report

Attention: Patch coverage is 46.15385% with 7 lines in your changes are missing coverage. Please review.

❗ No coverage uploaded for pull request base (mlx@722883a). Click here to learn what that means.

Files	Patch %	Lines
keras/src/backend/mlx/core.py	0.00%	5 Missing ⚠️
keras/src/backend/common/variables.py	0.00%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@          Coverage Diff           @@
##             mlx   #19619   +/-   ##
======================================
  Coverage       ?   69.12%           
======================================
  Files          ?      506           
  Lines          ?    45944           
  Branches       ?     8499           
======================================
  Hits           ?    31757           
  Misses         ?    12534           
  Partials       ?     1653

Flag	Coverage Δ
keras	`69.04% <46.15%> (?)`
keras-jax	`59.51% <46.15%> (?)`
keras-numpy	`53.85% <46.15%> (?)`
keras-tensorflow	`60.65% <46.15%> (?)`

Flags with carried forward coverage won't be shown. Click here to find out more.

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

…st and channels_last data formats

…lity and fix validation checks

…checks

…st and channels_last data formats

…keras_model function

…st and channels_last data formats

…t and channels_last data formats

…t and channels_last data formats v2

…t and channels_last data formats consistently

…as default value

…stent formatting for Conv2D layers

…irst and channels_last data formats consistently

…nput_shape parameter in Conv2D layer

…put_shape parameter in Conv2D layer

…put_shape parameter from Conv2D layer

…ions

…ging print statements

…ions

lkarthee · 2024-05-04T06:32:50Z

keras/src/backend/mlx/core.py

+    if isinstance(x, mx.array):
+        if x.dtype == mx.int64:
+            x = x.astype(mx.int32)
+        elif x.dtype == mx.float64:


there is no mx.float64 ?

lkarthee · 2024-05-04T06:49:36Z

keras/src/backend/mlx/core.py

@@ -70,9 +70,11 @@ def convert_to_tensor(x, dtype=None, sparse=None):
            return x.value.astype(mlx_dtype)
        return x.value

-    if isinstance(x, np.ndarray):


If we remove this cast, mlx crashes when running some test cases on my local machine. I logged a bug on mlx repo (see mlx backend issue).

It did not crash on CI tests on this pr because it is not being run on macOS ? Am really confused as I thought mlx runs only on macOS.

https://github.com/keras-team/keras/actions/runs/8907542811/job/24461598952#step:1:2

2024-05-01T09:07:10.5470064Z Current runner version: '2.316.0' 2024-05-01T09:07:10.5494922Z ##[group]Operating System 2024-05-01T09:07:10.5495654Z Ubuntu 2024-05-01T09:07:10.5496034Z 22.04.4 2024-05-01T09:07:10.5496669Z LTS 2024-05-01T09:07:10.5497089Z ##[endgroup] 2024-05-01T09:07:10.5497521Z ##[group]Runner Image 2024-05-01T09:07:10.5498174Z Image: ubuntu-22.04 2024-05-01T09:07:10.5498671Z Version: 20240422.1.0 2024-05-01T09:07:10.5499877Z Included Software: https://github.com/actions/runner-images/blob/ubuntu22/20240422.1/images/ubuntu/Ubuntu2204-Readme.md 2024-05-01T09:07:10.5501763Z Image Release: https://github.com/actions/runner-images/releases/tag/ubuntu22%2F20240422.1

Implement convolution operation for MLX backend

9648968

google-ml-butler bot added the size:M label Apr 26, 2024

google-ml-butler bot assigned gbaned Apr 26, 2024

Faisal-Alsrheed changed the title ~~[WIP] Implement convolution operation for MLX backend~~ [WIP] Implement backend/mlx/nn.py:conv Apr 26, 2024

Faisal-Alsrheed added 25 commits April 26, 2024 06:34

Fix dilation_rate standardization in conv function

e73ec18

Fix validation check for input and kernel channels in conv function

804a7be

Refactor validation check for input and kernel channels in conv function

00160e7

Refactor conv function to improve readability and fix validation checks

f1bb716

Refactor _transpose_conv_kernel function to support both channels_fir…

55fbace

…st and channels_last data formats

Refactor _transpose_conv_kernel and conv functions to improve readabi…

365f48e

…lity and fix validation checks

Refactor numerical_test.py to support MLX backend and fix validation …

894c5d5

…checks

Refactor conv function

1b2f2bf

Fix validation check for input and kernel channels in conv function

437b7a0

Refactor _transpose_conv_kernel function to support both channels_fir…

5a62e65

…st and channels_last data formats

Refactor numerical_test.py to include data_format parameter in build_…

a35a56c

…keras_model function

Refactor _transpose_conv_kernel function to support both channels_fir…

f6278f0

…st and channels_last data formats

Refactor _transpose_conv_kernel function to handle both channels_firs…

4372351

…t and channels_last data formats

Refactor _transpose_conv_kernel function to handle both channels_firs…

9549880

…t and channels_last data formats v2

Refactor _transpose_conv_kernel function to handle both channels_firs…

6514899

…t and channels_last data formats consistently

Refactor data_format parameter in conv function to use channels_last …

682072e

…as default value

Refactor build_keras_model function in numerical_test.py to use consi…

6d7f5dd

…stent formatting for Conv2D layers

Refactor conv function to include support for groups parameter

b96047e

Refactor _transpose_spatial_inputs function to handle both channels_f…

9e25cb4

…irst and channels_last data formats consistently

Refactor build_keras_model function in numerical_test.py to include i…

9f0aa30

…nput_shape parameter in Conv2D layer

Refactor build_keras_model function in numerical_test.py to update in…

e63913d

…put_shape parameter in Conv2D layer

Refactor build_keras_model function in numerical_test.py to remove in…

a6941c1

…put_shape parameter from Conv2D layer

fix

b9d116d

fix

4390780

fix

5fa46eb

Faisal-Alsrheed added 7 commits April 28, 2024 14:17

Refactor dtype conversion in to_mlx_dtype and convert_to_tensor funct…

fb27c5b

…ions

Refactor dtype conversion in convert_to_tensor function and add debug…

8b9331a

…ging print statements

Refactor dtype conversion in convert_to_tensor and to_mlx_dtype funct…

6dc86da

…ions

No /workspaces/keras/integration_tests/numerical_test.py

690e243

add debugging print statements

77314fe

fix convert_to_tensor

8e040db

Refactor dtype conversion in convert_to_tensor and to_mlx_dtype funct…

0668b46

…ions

Faisal-Alsrheed closed this Apr 29, 2024

Faisal-Alsrheed reopened this Apr 29, 2024

Faisal-Alsrheed added 9 commits April 29, 2024 19:31

Refactor dtype handle MLX

59c9af9

Refactor dtype handle MLX

19171b7

Refactor dtype handle MLX

44a33ec

Refactor dtype handling and add MLX backend support

62be51b

Refactor dtype handling and add explicit handling for float8 dtypes

7f75c3a

comments in result_type

82776dd

Add core test file for mlx

fc78a65

Refactor dtype handling

02dc92a

no numerical_test.py

de83556

Faisal-Alsrheed changed the title ~~[WIP] Implement backend/mlx/nn.py:conv~~ [WIP] Fix backend/mlx/core file and backend/common/dtypes for MLX Apr 30, 2024

Faisal-Alsrheed changed the title ~~[WIP] Fix backend/mlx/core file and backend/common/dtypes for MLX~~ [WIP] Fix backend/mlx/core and backend/common/dtypes for MLX Apr 30, 2024

improve numerical_test.py and standardize_dtype

213a7fc

Faisal-Alsrheed changed the title ~~[WIP] Fix backend/mlx/core and backend/common/dtypes for MLX~~ [WIP] Fix backend/mlx/core and backend/common/dtypes for MLX + Improve integration_tests/numerical_test.py May 1, 2024

Faisal-Alsrheed added 3 commits May 1, 2024 07:48

Add DEBUGGING = True

2e40687

Refactor numerical_test.py and variables.py

c3bfdcb

Refactor dtype handling and add MLX backend support

4eb82b3

Faisal-Alsrheed mentioned this pull request May 1, 2024

mlx - implement segment_sum and segment_max #19652

Open

Faisal-Alsrheed marked this pull request as ready for review May 3, 2024 06:08

Faisal-Alsrheed changed the title ~~[WIP] Fix backend/mlx/core and backend/common/dtypes for MLX + Improve integration_tests/numerical_test.py~~ Fix backend/mlx/core and backend/common/dtypes for MLX + Improve integration_tests/numerical_test.py May 3, 2024

lkarthee reviewed May 4, 2024

View reviewed changes

gbaned added this to Assigned Reviewer in PR Queue via automation May 7, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix `backend/mlx/core` and `backend/common/dtypes` for MLX + Improve `integration_tests/numerical_test.py` #19619

Fix `backend/mlx/core` and `backend/common/dtypes` for MLX + Improve `integration_tests/numerical_test.py` #19619

Faisal-Alsrheed commented Apr 26, 2024 •

edited

codecov-commenter commented Apr 26, 2024 •

edited

lkarthee May 4, 2024 •

edited

lkarthee May 4, 2024 •

edited

Fix backend/mlx/core and backend/common/dtypes for MLX + Improve integration_tests/numerical_test.py #19619

Are you sure you want to change the base?

Fix backend/mlx/core and backend/common/dtypes for MLX + Improve integration_tests/numerical_test.py #19619

Conversation

Faisal-Alsrheed commented Apr 26, 2024 • edited

codecov-commenter commented Apr 26, 2024 • edited

Codecov Report

lkarthee May 4, 2024 • edited

Choose a reason for hiding this comment

lkarthee May 4, 2024 • edited

Choose a reason for hiding this comment

Fix `backend/mlx/core` and `backend/common/dtypes` for MLX + Improve `integration_tests/numerical_test.py` #19619

Fix `backend/mlx/core` and `backend/common/dtypes` for MLX + Improve `integration_tests/numerical_test.py` #19619

Faisal-Alsrheed commented Apr 26, 2024 •

edited

codecov-commenter commented Apr 26, 2024 •

edited

lkarthee May 4, 2024 •

edited

lkarthee May 4, 2024 •

edited