v0.7.0

NOBLES5E released this 16 Aug 11:29

· 596 commits to master since this release

Bug Fixes

Autotune api conflict (#131)

Features

Add low precision decentralized algorithm (#103)
Add all communication primitives such as send recv to communication module (#128)
Make full precision decentralized op stateless (#126)
Support nccl 2.10 ReduceOp.AVG (#149)
Add support for reporting tensor completion order (#146)

Assets 2