Tag
This paper replicates a study on AI efficiency assessment using FLOPs, validates that raw FLOPs are not a suitable metric for execution time on newer hardware, and emphasizes the need for complete replication packages in research.
A systematic review and empirical study comparing carbon footprints of deep learning models, examining Green AI techniques and measurement tools, with findings that training dominates emissions and that larger architectures do not always yield proportionate accuracy gains.
This paper presents an analytically structured, empirically calibrated methodology for estimating LLM inference energy on NVIDIA H100 GPUs without direct measurement, separating prefill and decoding phases and decomposing energy into compute, parameter-access, KV-cache write, and attention-read components.
This survey comprehensively reviews resource-efficient architectures and hardware-software co-design for green AI, covering efficient model construction, training/deployment strategies, and sustainable hardware, aiming to guide sustainable large model development.
This paper proposes A-LEMS, a framework that redefines AI energy accounting from per-inference to Energy per Successful Goal (EpG), and introduces the Orchestration Overhead Index (OOI) to measure energy costs of multi-step orchestration in agentic systems. Empirical results show agentic workflows consume 4.33× higher mean energy per goal than linear baselines, but OOI can invert for tool-augmented tasks, demonstrating goal-level accounting is necessary.
This paper presents an end-to-end energy accounting framework for LLM distillation pipelines, measuring stage-wise energy costs and constructing energy-quality Pareto frontiers to reveal previously ignored teacher-side costs.