v0.7.0

Changed

• Bumped XLA version (#83)
• Renamed the recognised XLA_TARGET "cuda120" to "cuda12" (#84)

Removed

• Dropped support for CUDA 11.8+, now 12.1+ is required (#84)

v0.6.0

Changed

• Bumped XLA version (#62)

v0.5.1

Changed

• Bumped the version requirement for CUDA 12 to CUDA 12.1 and cuDNN 8.9 (#54)

v0.5.0

Added

• Support for custom http headers (#44)
• Support for CUDA 12

Changed

• Migrated to OpenXLA source code (#45)

Removed

• Dropped precompiled binary for CUDA 11.1 and CUDA 11.4
• Dropped precompiled binary for Linux musl

v0.4.4

Added

• Sorting library (#39)

v0.4.3

Fixed

• Building with XLA_BUILD (regression from v0.4.2) (#33)

v0.4.2

Added

• Precompiled binaries for Linux musl (#31)

Fixed

• Partially fixed building for ROCm, see notes (#30)

v0.4.1

Added

• Precompiled binaries for CUDA 11.4+ (cuDNN 8.2+) and CUDA 11.8+ (cuDNN 8.6+) (#27)

Changed

• Precompiled binaries to assume glibc 31+ (#27)

v0.4.0

Changed

• Bumped XLA (Tensorflow) version to 2.11.0 (#25)

v0.3.0

Changed

• Bumped XLA (Tensorflow) version to 2.8.0 (#15)

Removed

• Dropped support for CUDA 10.2 and 11.0, now 11.1+ is required (#17)