Ray: A Distributed Framework for Emerging AI Applications
About
The next generation of AI applications will continuously interact with the environment and learn from these interactions. These applications impose new and demanding systems requirements, both in terms of performance and flexibility. In this paper, we consider these requirements and present Ray---a distributed system to address them. Ray implements a unified interface that can express both task-parallel and actor-based computations, supported by a single dynamic execution engine. To meet the performance requirements, Ray employs a distributed scheduler and a distributed and fault-tolerant store to manage the system's control state. In our experiments, we demonstrate scaling beyond 1.8 million tasks per second and better performance than existing specialized systems for several challenging reinforcement learning applications.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Reinforcement Learning | Procgen (test) | BigFish Return8.2 | 21 | |
| Reinforcement Learning | Procgen BigFish 1.0 (train) | Mean Train Performance20.8 | 6 | |
| Reinforcement Learning | Procgen CoinRun 1.0 (train) | Mean Train Performance10 | 6 | |
| Reinforcement Learning | Procgen FruitBot 1.0 (train) | Mean Train Performance32.2 | 6 | |
| Reinforcement Learning | Procgen StarPilot 1.0 (train) | Mean Train Performance44.1 | 6 | |
| Reinforcement Learning | Procgen CaveFlyer 1.0 (train) | Mean Performance (Train)7.3 | 6 | |
| Reinforcement Learning | Procgen Jumper 1.0 (train levels) | Mean Train Performance9 | 6 | |
| Reinforcement Learning | Procgen Leaper 1.0 (train) | Mean Train Performance6.9 | 6 |