QuaRL | source framework | Reinforcement Learning library

QuaRL is a Python library typically used in Institutions, Learning, Education, Artificial Intelligence, Reinforcement Learning, Deep Learning, Pytorch, Tensorflow applications. QuaRL has no bugs, it has no vulnerabilities and it has low support. However QuaRL build file is not available. You can download it from GitHub.

Deep reinforcement learning has achieved significant milestones, however, the computational demands of reinforcement learning training and inference remain substantial. Using quantization techniques such as Post Training Quantization and Quantization Aware Training, a well-known technique in reducing computation costs, we perform a systematic study of Reinforcement Learning Algorithms such as A2C, DDPG, DQN, PPO and D4PG on common environments. Motivated by the effectiveness of PTQ, we propose ActorQ, a quantized actor-learner distributed training system that runs learners in full precision and actors in quantized precision (fp16, int8). We demonstrated end-to-end speedups of 1.5x - 2.5x in reinforcement learning training with no loss in reward. Further, we breakdown the various runtime costs in distributed reinforcement learning training and show the effects of quantization on each. The framework currently support the following environments, RL algorithms and quantization methods. Read the paper here for more information: