NVIDIA AI has announced the integration of FP8 support in its open-source library, NVIDIA NeMo RL, to enhance reinforcement learning (RL) training. The addition of low-precision FP8 format aims to accelerate computational workloads, improving the speed of RL tasks by 1.48 times on the Qwen3-8B-Base model. This enhancement is expected to facilitate faster iteration cycles for agent tool usage and multi-step tasks, boosting efficiency in RL applications.