Skip to content

Intel® oneCCL Bindings for Pytorch* v2.1.300+xpu release note
Compare
Choose a tag to compare
@zhangxiaoli73 zhangxiaoli73 released this 26 Apr 12:25
· 1 commit to ccl_torch2.1.300+xpu since this release
v2.1.300+xpu
1053f13
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Learn about vigilant mode.

Features include:

  • Extend a prototype feature enabled by TORCH_LLM_ALLREDUCE=1 to provide better scale-up performance by enabling optimized collectives such as allreduce, allgather, reducescatter algorithms in Intel® oneCCL. This feature requires XeLink enabled for cross-cards communication.
  • Enable a set of coalesced primitives in CCL backend, including allreduce_into_tensor_coalesced, allgather_into_tensor_coalesced , reduce_scatter_tensor_coalesced and _broadcast_coalesced.
Assets 2
Loading