Position: LLM Inference Should Be Evaluated as Energy-to-Token Production
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910211569090560 |
|---|---|
| author | Liu, Xiang Yuan, Shimiao Tang, Zhenheng Dong, Peijie Zhao, Kaiyong Wang, Qiang Li, Bo Chu, Xiaowen |
| author_facet | Liu, Xiang Yuan, Shimiao Tang, Zhenheng Dong, Peijie Zhao, Kaiyong Wang, Qiang Li, Bo Chu, Xiaowen |
| contents | LLM inference is still evaluated mainly as a model or software problem: accuracy, latency, throughput, and hardware utilization. This is incomplete. At deployment scale, the relevant output is a quality-conditioned token produced under joint constraints from effective compute, delivered data-center power, cooling capacity, PUE, and utilization.
We argue that the ML community should treat inference as \emph{energy-to-token production}. We formalize this view with a dimensionally consistent Token Production Function in which token rate is bounded by both compute-per-token and energy-per-token ceilings. Listed API prices vary by over an order of magnitude across providers, but we use price dispersion only as directional motivation, not as causal evidence of marginal cost. The core physical question is instead: under fixed quality and service targets, when does the binding constraint move from theoretical peak compute toward delivered power, cooling, and operational efficiency?
Under this framing, system optimizations -- latent KV-cache compression, sparse or heavily compressed attention, quantization, routing, and difficulty-adaptive reasoning -- are not merely local engineering tricks. They are energy-to-token levers because they reduce FLOPs/token, joules/token, memory traffic, or utilization losses under fixed $(q^{*},s^{*})$. We therefore call for inference papers and benchmarks to report Joules/token, active binding constraint, PUE-adjusted delivered power, and utilization-adjusted token output alongside accuracy and latency. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_11733 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Position: LLM Inference Should Be Evaluated as Energy-to-Token Production Liu, Xiang Yuan, Shimiao Tang, Zhenheng Dong, Peijie Zhao, Kaiyong Wang, Qiang Li, Bo Chu, Xiaowen Computational Engineering, Finance, and Science Distributed, Parallel, and Cluster Computing LLM inference is still evaluated mainly as a model or software problem: accuracy, latency, throughput, and hardware utilization. This is incomplete. At deployment scale, the relevant output is a quality-conditioned token produced under joint constraints from effective compute, delivered data-center power, cooling capacity, PUE, and utilization. We argue that the ML community should treat inference as \emph{energy-to-token production}. We formalize this view with a dimensionally consistent Token Production Function in which token rate is bounded by both compute-per-token and energy-per-token ceilings. Listed API prices vary by over an order of magnitude across providers, but we use price dispersion only as directional motivation, not as causal evidence of marginal cost. The core physical question is instead: under fixed quality and service targets, when does the binding constraint move from theoretical peak compute toward delivered power, cooling, and operational efficiency? Under this framing, system optimizations -- latent KV-cache compression, sparse or heavily compressed attention, quantization, routing, and difficulty-adaptive reasoning -- are not merely local engineering tricks. They are energy-to-token levers because they reduce FLOPs/token, joules/token, memory traffic, or utilization losses under fixed $(q^{*},s^{*})$. We therefore call for inference papers and benchmarks to report Joules/token, active binding constraint, PUE-adjusted delivered power, and utilization-adjusted token output alongside accuracy and latency. |
| title | Position: LLM Inference Should Be Evaluated as Energy-to-Token Production |
| topic | Computational Engineering, Finance, and Science Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2605.11733 |