Repository for hosting material and discussions for the 2021 IDRIS GPU hackathon
In this hackathon we will be working on 3 layers:
- Horovod Fork: Interface to NCCL comms from TensorFlow
- Mesh TensorFlow Fork: SPMD Mesh implementation based on Horovod
- FlowPM: Distributed PM solver based on Mesh TensorFlow
To get started with the hackathon checkout GETTING_STARTED.md