Skip to content

v0.8.0

Choose a tag to compare

@aoyulong aoyulong released this 04 Jan 13:17
· 31 commits to main since this release
560d2f8
  • Enables intra-node zero-copy to improve data transfer efficiency for small messages.
  • Supports a naive AllReduce implementation in uniRunner mode using a CPU-centric, device-assisted algorithm.
  • Adds one-sided communication primitives via the new APIs flagcxHeteroPut and flagcxHeteroPutSignal.