ML.ENERGY Benchmark
收藏资源简介:
ML.ENERGY Benchmark是一个用于测量生成式AI模型在真实服务环境中的推理能耗的工具和数据集。该数据集由密歇根大学的研究团队创建,旨在帮助理解和优化生成式AI服务的能耗。数据集包含了40种广泛使用的模型架构在6个不同任务中的能耗测量结果,并提供了自动化优化建议,可以在不改变模型计算内容的情况下实现显著的能耗节约。该数据集是开源的,可以轻松扩展到各种定制模型和应用场景。
The ML.ENERGY Benchmark is a tool and dataset for measuring the inference energy consumption of generative AI models in real-world service environments. This dataset was developed by a research team from the University of Michigan, with the goal of facilitating the understanding and optimization of energy consumption for generative AI services. It contains energy consumption measurement results of 40 widely adopted model architectures across 6 distinct tasks, and provides automated optimization recommendations that can achieve substantial energy savings without altering the computational logic of the models. This dataset is open-source and can be readily extended to various custom models and application scenarios.




