Relay is a high level intermediate representation for the TVM framework. The goal of Relay is to replace old computation graph based IRs such as NNVM with a more expressive IR that can be effectively optimized for many targets.
Relay is still in active research and development, but you can find the initial alpha version in TVM’s master branch.
Look forward to more documentation and papers in the coming weeks and months.
For now you can read our workshop paper from earlier in the year.