Transparent Screening for LLM Inference and Training Impacts

Arnault Pachot; Thierry Petit

cs.LGMar 23, 2026

Transparent Screening for LLM Inference and Training Impacts

Authors: Arnault Pachot, Thierry Petit

Organizations: Emotia, Paris, France

Abstract

This paper presents a transparent screening framework for estimating inference and training impacts of current large language models under limited observability. The framework converts natural-language application descriptions into bounded environmental estimates and supports a comparative online observatory of current market models. Rather than claiming direct measurement for opaque proprietary services, it provides an auditable, source-linked proxy methodology designed to improve comparability, transparency, and reproducibility.

Figures & tables

Case	Energy	Carbon	Interpretation
Ministral 3B, one standardized request	0.0056 Wh	0.0002 gCO 2 e	Small hybrid model under a French provider-country proxy
GPT-5 mini, one standardized request	0.1706 Wh	0.0657 gCO 2 e	Medium proprietary hosted model under a US provider-country proxy
Claude Opus 4.1, one standardized request	2.9922 Wh	1.1520 gCO 2 e	Very large proprietary estimate illustrating the steep growth of screening orders of magnitude
Support chatbot, Ministral 8B, 20,000 conversations/month	2.38 kWh/year	96 gCO 2 e/year	Low-carbon annual result despite high usage because the retained electricity factor is favorable
Retrieval assistant, GPT-5 mini, 4,000 uses/month	12.31 kWh/year	4.74 kgCO 2 e/year	Higher annualized result driven by hosted-service assumptions and US electricity contextualization

Table 1: Illustrative outputs from the current observatory and application estimator. These values are screening estimates, not audited declarations.

Model	Inference Wh/request	Inference gCO 2 e/request	Training GWh	Interpretation
Claude Opus 4.1	2.9922	1.1520	125.63	Upper-end proprietary model with very large screening values for both inference and training
GPT-5.2	2.7897	1.0740	101.76	Frontier proprietary model illustrating the high end of current screening estimates
Gemini 2.5 Pro	0.3601	0.1387	1.26	Upper-mid-range model with markedly lower screening values than the top tier
GPT-5 mini	0.1706	0.0657	0.28	Medium model representative of lighter hosted deployments
Llama 3.1 70B	0.1043	0.0402	0.16	Large open-weight model with moderate inference estimates
GPT-4o mini	0.0155	0.0060	0.0027	Lightweight hosted model with low per-request screening impacts

Table 2: Representative subset of observatory results spanning the current market-model catalog. Inference values correspond to the central prompt-level proxy retained by the observatory, expressed per standardized request. Training values are multi-factor screening estimates; additional decimals are shown for small training values to avoid false zeros from rounding.

LLM Auditing

Large language model auditing tests models and deployed systems for bias, safety, privacy, reliability, and provenance using behavioral and internal evidence.

See all 639 papers

LLM-FACETS: A Privacy-Preserving Framework for Evaluating LLM Transparency and Accountability
May 29, 2026Tom Lucas, Alessio Buscemi, Alfredo Capozucca +2LLM Evaluation AI Accountability
Luxembourg Institute of Science and Technology (LIST), Luxembourg, University of Luxembourg, Luxembourg, and Scriptor Artis, France · Luxembourg Institute of Science and Technology (LIST), Luxembourg · University of Luxembourg, Luxembourg +1
Privacy-Preserving Mechanisms Enable Cheap Verifiable Inference of LLMs
Feb 19, 2026Arka Pal, Louai Zahran, William Gvozdjak +2Verifiable Outsourced Computation LLM Inference
Ritual · MIT · Columbia University
IMMACULATE: A Practical LLM Auditing Framework via Verifiable Computation
Feb 26, 2026Yanpei Guo, Wenjie Qu, Linyu Wu +7Verifiable Outsourced Computation LLM Auditing
National University of Singapore · Nanyang Technological University · Independent Researcher +1

May 29, 2026J/K move · Enter open · S save

LLM-FACETS: A Privacy-Preserving Framework for Evaluating LLM Transparency and Accountability

Tom Lucas, Alessio Buscemi, Alfredo Capozucca, German Castignani+1

Luxembourg Institute of Science and Technology (LIST), Luxembourg, University of Luxembourg, Luxembourg, and Scriptor Artis, France · Luxembourg Institute of Science and Technology (LIST), Luxembourg · University of Luxembourg, Luxembourg+1

LLM Evaluation AI Accountability LLM Auditing LLM Reliability

Open paper AskPDF arXiv

Assessing whether Large Language Models outputs are factually grounded, epistemically calibrated, and methodologically reproducible is a prerequisite for responsible AI deployment. Yet auditing LLMs remains inaccessible to non-technical practitioners: existing tools require programming expertise and non-trivial environment setup, and cloud-hosted platforms transmit evaluation data to external services, creating barriers for domain experts and compliance officers legally responsible for AI oversight. We introduce LLM-FACETS (LLM FActuality Cross-EvaluaTion System): an open-source framework with a browser-accessible interface and a plugin architecture, structured around three practitioner profiles (technical experts, domain experts, compliance officers) that mirror the stakeholder categories identified in the EU AI Act and the NIST AI Risk Management Framework. The architecture makes data flows explicit: deterministic metrics (BLEU, ROUGE, BERTScore) run entirely within the self-hosted server with no outbound transmission; LLM-judge metrics contact external APIs explicitly, with users retaining full credential control. The framework operationalizes transparency through three mechanisms: token-level log-probability visualization for epistemic uncertainty, multi-judge consensus to mitigate judge bias, and RAG Triad metrics (Faithfulness, Answer Relevance, Context Relevance) to detect and localize hallucinations. A plugin architecture allows any new metric or dataset to be integrated without modifying the evaluation pipeline. The open-source implementation enables cross-checking across multiple metrics targeting the same property, ensuring reproducibility and decoupling AI accountability from the teams building the systems assessed. We verify the framework through cross-validation of 18 metric implementations against canonical reference libraries.

Abstract

LLM Auditing

Explore similar work

LLM-FACETS: A Privacy-Preserving Framework for Evaluating LLM Transparency and Accountability

Privacy-Preserving Mechanisms Enable Cheap Verifiable Inference of LLMs

IMMACULATE: A Practical LLM Auditing Framework via Verifiable Computation