Saved in:
Bibliographic Details
Main Authors: Shi, Tianyao, Ding, Yi
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.27480
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917542108332032
author Shi, Tianyao
Ding, Yi
author_facet Shi, Tianyao
Ding, Yi
contents Large language model (LLM) serving creates environmental impacts beyond carbon and water, including ecosystem damage through biodiversity-related pathways. We present BIRDS, a framework for Biodiversity Impact of Request-Driven LLM Serving. BIRDS defines request-level functional units, quantifies operational and embodied biodiversity impact, and introduces Quality-Normalized Biodiversity Impact (QNBI) to jointly analyze ecological impact and response quality. Across diverse workloads, models, GPUs, and regions, BIRDS reveals that biodiversity impact accumulates at scale and exposes actionable quality-aware serving tradeoffs.
format Preprint
id arxiv_https___arxiv_org_abs_2605_27480
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle BIRDS: Characterizing and Understanding Biodiversity Impact of Large Language Model Serving
Shi, Tianyao
Ding, Yi
Other Quantitative Biology
Artificial Intelligence
Computers and Society
Large language model (LLM) serving creates environmental impacts beyond carbon and water, including ecosystem damage through biodiversity-related pathways. We present BIRDS, a framework for Biodiversity Impact of Request-Driven LLM Serving. BIRDS defines request-level functional units, quantifies operational and embodied biodiversity impact, and introduces Quality-Normalized Biodiversity Impact (QNBI) to jointly analyze ecological impact and response quality. Across diverse workloads, models, GPUs, and regions, BIRDS reveals that biodiversity impact accumulates at scale and exposes actionable quality-aware serving tradeoffs.
title BIRDS: Characterizing and Understanding Biodiversity Impact of Large Language Model Serving
topic Other Quantitative Biology
Artificial Intelligence
Computers and Society
url https://arxiv.org/abs/2605.27480