Skip to content

TfLite Mul adds missing datatype support - #45

Merged
RahulSundarMCW merged 2 commits into
tflite_mul_missing_datatypefrom
nithesh_tflite_mul_op_missing_datatype
Apr 1, 2025
Merged

RahulSundarMCW merged 2 commits into
tflite_mul_missing_datatypefrom
nithesh_tflite_mul_op_missing_datatype

Conversation

@nitheshsrikanth-mcw

Copy link
Copy Markdown
Collaborator
  • Adds f16,bf16 for mul
  • Adds f16,bf16 unit tests

@nitheshsrikanth-mcw nitheshsrikanth-mcw self-assigned this Jul 29, 2024
Comment thread tensorflow/lite/kernels/internal/types.h Outdated
Comment thread tensorflow/lite/kernels/mul.cc Outdated
@nitheshsrikanth-mcw
nitheshsrikanth-mcw force-pushed the nithesh_tflite_mul_op_missing_datatype branch from 80a5e85 to d78e99d Compare August 16, 2024 09:21
@nitheshsrikanth-mcw
nitheshsrikanth-mcw force-pushed the nithesh_tflite_mul_op_missing_datatype branch 3 times, most recently from 9d95900 to 7ab407c Compare August 26, 2024 11:43
@nishantsarda-mcw

nishantsarda-mcw commented Aug 26, 2024 •

Copy link
Copy Markdown
Collaborator

Hi Luke, I wanted to discuss the current status of mul test file, which has now exceeded 780 lines due to the addition of tests for f16 and bf16 to cover the same behavior as f32. Given the significant increase in file size, I'm considering whether it's necessary to test all f32 scenarios for f16 and bf16, or if we might focus on a more selective set of tests.
or if we are to have all the set of tests then would it be possible to split the PR into two separate submissions? One could address the f16 tests and the other the bf16 tests.
Also, Can we squash the above commits into single commit, so that it will streamline the review process?

@amrinfathima-mcw
amrinfathima-mcw force-pushed the nithesh_tflite_mul_op_missing_datatype branch from 7ab407c to 6967706 Compare August 26, 2024 13:15
 - Adds f16,bf16 for mul
 - Adds f16,bf16  unit tests

Addressed review comments and included optimized versions of Mul for bf16/f16

-Adds optimized versions for bf16/f16 mul operation
-Addressed review comments
-Included the EIGEN_TFLITE flag and other changes to resolve CONV Error
@nitheshsrikanth-mcw
nitheshsrikanth-mcw force-pushed the nithesh_tflite_mul_op_missing_datatype branch from 6967706 to 841f474 Compare October 24, 2024 03:57
@nitheshsrikanth-mcw

Copy link
Copy Markdown
Collaborator Author

Hi luke, I have squashed the commits and addressed the changes.

Comment thread tensorflow/lite/kernels/mul.cc
Comment thread tensorflow/lite/kernels/mul.cc
@RahulSundarMCW
RahulSundarMCW merged commit 3487fe8 into tflite_mul_missing_datatype Apr 1, 2025
RahulSundarMCW added a commit that referenced this pull request Apr 1, 2025
* TfLite Mul add missing datatype support
 - Adds f16,bf16 for mul
 - Adds f16,bf16  unit tests

Addressed review comments and included optimized versions of Mul for bf16/f16

-Adds optimized versions for bf16/f16 mul operation
-Addressed review comments
-Included the EIGEN_TFLITE flag and other changes to resolve CONV Error

* Include non-quantized int8 type support

---------

Co-authored-by: nitheshsrikanth-mcw <[email protected]>
RahulSundarMCW added a commit that referenced this pull request May 7, 2025
* feat: add missing type support for tflite_add (#107) (#108)

* TfLite add missing datatype support

 - Adds bf16, f16 support sub function
 - Adds bf16, f16 unit tests

* Add non-quantized int8 type support and tests

* Include check for broadcasting

* Tflite sign missing datatype support (#37) (#109)

* Tflite sign missing datatype support

-Adds i16,f16,bf16 for tflite sign
-Adds i16,f16,bf16 tflite sign unit tests

* Add non-quantized int8 type support

---------

Co-authored-by: swatheesh-mcw <[email protected]>

* TfLite_sub_op add missing datatype support (#35) (#110)

* TfLite sub missing datatype support

 - Adds bf16, f16 support sub function
 - Adds bf16, f16 unit tests

* Include non-quantized int8&int16 type support

* Tflite ceil missing datatypes support (#51) (#111)

* Tflite ceil missing datatypes support
-Adds bf16,f16,i8,i16,i32 for tflite ceil operations
-Adds bf16,f16,i8,i16,i32 ceil unit tests

* Add unquantized int8 support

---------

Co-authored-by: abhinav <[email protected]>

* TfLite Mul adds missing datatype support (#45)

* TfLite Mul add missing datatype support
 - Adds f16,bf16 for mul
 - Adds f16,bf16  unit tests

Addressed review comments and included optimized versions of Mul for bf16/f16

-Adds optimized versions for bf16/f16 mul operation
-Addressed review comments
-Included the EIGEN_TFLITE flag and other changes to resolve CONV Error

* Include non-quantized int8 type support

---------

Co-authored-by: RahulSudarMCW <[email protected]>

---------

Co-authored-by: swatheesh-mcw <[email protected]>
Co-authored-by: abhinav <[email protected]>
Co-authored-by: nitheshsrikanth-mcw <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants