Skip to content

Conversation

metax666
Copy link
Collaborator

No description provided.

zhang-chenyi and others added 12 commits September 4, 2025 14:55
[Metax] fix fail test when backend is mack
* [Metax_change_ut]

* fix sum&collect_fpn_proposals op register

* modify profile

* [Metax] fix paddle bug replace 'MoeGradDispatchKernel' to 'MoeGateDispatchKernel'

* [Metax] register bce_loss_grad & bce_loss & index_add_grad kernels

* [Metax] con2d_grad use gpudnn

* blas handle support

* [Metax] register some kernels & update CMakeLists

* [Metax] fix metax unittest fail

* [Metax] add group_norm & label_smooth kernel and update matmul kernel

* [Metax] fix rmsprop kernel register and add meshgrid & meshgrid_grad kernel register

* add test

* add test

* [test]  chang the logic of workspace_host in cholesky_kernel_register

alloc(cpuplace,size), test pass
alloc(cpuplace, size, stream), crash

* [Metax] fix compile fail

* Revert "[Metax] fix compile fail"

This reverts commit 83bc87f.

* [Metax] fix compile fail by 'conv_transpose_grad_kernel_impl.h'

* [Metax]fix bug and add qr lstsq logsoftmax

* [Metax] con2d_grad use gpudnn

* [Metax]fix bug and add qr lstsq logsoftmax

* [Metax] change_patch

* [Metax] update unit test CMakeLists.txt

* [Metax] update unit test CMakeLists.txt

* [feature] add unique_consecutive kernel

* [metax] add some kernel

* [metax] add some kernel

* [Metax] register baddbmm kernel & update blas api

* [Metax] register baddbmm kernel & update blas api

* [Metax] register deformable_conv kernel & fix 'ModulatedDeformableCol2imCoord' symbol undefined

* [feature]  add add unique_consecutive kernel.cu

* [fix] fix some test case due to missing op register

* [fix]  fix some fail text

* [metax]fix lu eigvalshsqueeze rnn kernel

* [metax]fix lu eigvalshsqueeze rnn kernel

* add and fix some kernels

* [Metax] register deformable_conv kernel & fix 'ModulatedDeformableCol2imCoord' symbol undefined

* [Metax] fix conflict

* [Metax] adapt to paddle-cpu-20250901 & resolve the issue of 'test_elementwise_mul_op_metax' failure

* [Metax] update repeat_interleave kernel & ignore max op test

* [metax]fix lu eigvalshsqueeze rnn kernel

* [metax] chang patch fix copy

* [metax] chang patch fix copy

* [Metax] update metax_gpu unit test

* [Metax] fix test CMakeList.txt

* [metax]change_cupti_and_fix_softmax

* [metax]change_patch

* [metax]change_patch

---------

Co-authored-by: Mingkun.Zhang <[email protected]>
Co-authored-by: metax666 <[email protected]>
Co-authored-by: jiaxinWang-metax <[email protected]>
Co-authored-by: MingkunZhang <[email protected]>
Co-authored-by: chezhang <[email protected]>
Co-authored-by: zhang-chenyi <[email protected]>
Co-authored-by: ZhouDuan <[email protected]>
…#9)

* [Metax] fix dgc & mklml compile product path problem

* [Metax] fix accuracy kernel & add test_accuracy_op_metax.py unit test

* [Metax] add mixed_vector fix & update change patch
* [Metax] fix dgc & mklml compile product path problem

* [Metax] fix accuracy kernel & add test_accuracy_op_metax.py unit test

* [Metax] add mixed_vector fix & update change patch

* [Metax] update metax_gpu CMakeLists.txt
* [metax] chang patch fix copy

* [metax] chang patch fix copy

* [Metax] update metax_gpu unit test

* [Metax] fix test CMakeList.txt

* [metax]change_cupti_and_fix_softmax

* [metax]change_patch

* [metax]change_patch

* [metax] updata_qr_kernel

* [metax] updata_qr_kernel

---------

Co-authored-by: Mingkun.Zhang <[email protected]>
Co-authored-by: metax666 <[email protected]>
Co-authored-by: jiaxinWang-metax <[email protected]>
Co-authored-by: MingkunZhang <[email protected]>
Co-authored-by: chezhang <[email protected]>
Co-authored-by: zhang-chenyi <[email protected]>
Co-authored-by: ZhouDuan <[email protected]>
* [Metax] fix illegal address access error in test_momentum_op
* [Metax] fix cufft and fix some blas kernel apply
Copy link

paddle-bot bot commented Sep 15, 2025

Thanks for your contribution!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants