Closed abdulfatir closed 2 months ago
Description of changes: This PR removes casting to fp32 for the cumsum operation and upgrades mlx to ~=0.10.0 which adds bf16 support for cumsum.
fp32
cumsum
mlx
~=0.10.0
bf16
Related: https://github.com/ml-explore/mlx/issues/959
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.
Description of changes: This PR removes casting to
fp32for thecumsumoperation and upgradesmlxto~=0.10.0which addsbf16support forcumsum.Related: https://github.com/ml-explore/mlx/issues/959
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.