swift-coreml-diffusers

Commit Graph

Author	SHA1	Message	Date
Pedro Cuenca	3a4e04ab1a	Decide attention and ANE on hardware capabilities. For chips with >= 8 performance cores, we select CPU+GPU (original attention). Otherwise we select CPU+ANE (split einsum). Some computers (M1 Pro, 16 core GPU) might yield slightly better performance using CPU+GPU+ANE with SPLIT_EINSUM.	2 years ago
Pedro Cuenca	8796695928	Advanced Settings: ANE (#28 ) * Preparing to allow users to override inference settings. * ANE setting. * Do not show advanced settings if ane is not available.	2 years ago
Pedro Cuenca	06ce6e56b5	macOS UI (#12 ) Native macOS UI Co-authored-by: Cyril Zakka <1841186+cyrilzakka@users.noreply.github.com>	2 years ago

Author

SHA1

Message

Date

Pedro Cuenca

3a4e04ab1a

Decide attention and ANE on hardware capabilities.

For chips with >= 8 performance cores, we select CPU+GPU (original
attention). Otherwise we select CPU+ANE (split einsum).

Some computers (M1 Pro, 16 core GPU) might yield slightly better
performance using CPU+GPU+ANE with SPLIT_EINSUM.

Pedro Cuenca

8796695928

Advanced Settings: ANE (#28 )

* Preparing to allow users to override inference settings.

* ANE setting.

* Do not show advanced settings if ane is not available.

Pedro Cuenca

06ce6e56b5

macOS UI (#12 )

Native macOS UI

Co-authored-by: Cyril Zakka <1841186+cyrilzakka@users.noreply.github.com>

3 Commits (65f6eaa57a0a8493bf1d2fb5cd9226d46a18f9cc)