This release includes an update to the DLPack fast C exchange API convention

What's Changed

• [DOCS] Include exchange api in the docs by @tqchen in #181
• Update exchange API to be capsule following new convention by @tqchen in #180

Full Changelog: v1.2...v1.3

This version intrduces the following new changes to the standard

• Add support for Trainium device in #168
• Enforce strides to be not null when ndim is nonzero in #178
• Support DLPACK C Functions for Speed Exchange and Stream Handling in #174

Full Changelog: v1.1...v1.2

This release mainly focuses on low-precision data type support that is increasingly relevant in latest ML workloads

• Added support for F8
• Added support for F4
• Added support for F6

This release tags DLPack v1.0 release.

• Introduce versioning DLManagedTensorVersioned
• flags about exported tensor

This release acompanies array api changes, see the note below

Starting Python array API standard v2023, a new max_version argument is added to __dlpack__ for the consumer to signal the producer the maximal supported DLPack version. Starting DLPack 1.0, the DLManagedTensorVersioned struct should be used and the existing DLManagedTensor struct is considered deprecated, though a library should try to support both during the transition period if possible.

This release tags DLPack v1.0 release candidate. Main changes include

• Introduce versioning DLManagedTensorVersioned
• flags about exported tensor

This release acompanies array api changes, see the note below

Starting Python array API standard v2023, a new max_version argument is added to __dlpack__ for the consumer to signal the producer the maximal supported DLPack version. Starting DLPack 1.0, the DLManagedTensorVersioned struct should be used and the existing DLManagedTensor struct is considered deprecated, though a library should try to support both during the transition period if possible.

This is last release before DLPack ABI change to add versioning

• Add kDLBool

This release adds the following fields

• Add kDLHexagon
• Add kDLOneAPI
• Add DLPACK_VERSION and DLPACK_ABI_VERSION

• Add kDLROCMHost
• Add kDLCUDAManaged

Rename enum names:

• kDLGPU -> kDLCUDA
• kDLCPUPinned -> kDLCUDAHost

The ABI is backward compatible, as it is only change of constant name, exchange can still happen between the new version and old version.

• OpaqueHandle type
• Complex support
• Rename DLContext -> DLDevice
  • DLTensor.ctx -> DLTensor.device
  • This requires dependent frameworks to upgrade the type name.
  • The ABI is backward compatible, as it is only change of constant name, exchange can still happen between the new version and old version.