Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Towards Efficient Multi-LLM Inference: Characterization and Analysis of LLM Routing and Hierarchical Techniques

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This survey argues that routing and cascading across multiple LLMs can cut inference cost dramatically while keeping quality, and proposes the Inference Efficiency Score to make such systems comparable.

desk verdict A useful, accurate survey of LLM routing and cascading whose only novel piece—the Inference Efficiency Score—is an uninstantiated sketch that should be either dropped or developed before publication. read the letter →

arxiv 2506.06579 v1 pith:ZDJW2D3G submitted 2025-06-06 cs.LG cs.AIcs.CLcs.DC

Adarsh Prasad Behera , Jaya Prakash Champati , Roberto Morabito , Sasu Tarkoma , James Gross This is my paper
classification cs.LGcs.AIcs.CLcs.DC
keywords LLMroutinghierarchicalinferencecascadingEfficiencyScorecost-performancetrade-offedgemodelselectionbenchmarks
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper is a survey that tries to establish a practical thesis: routing and hierarchical inference (HI) are effective, general strategies for cutting the cost of serving LLM queries, not niche tricks. Routing sends each query to the single best-suited model, while HI starts with a small, cheap model and escalates to larger ones only when confidence is low, and the reviewed systems report savings of roughly 50–98% in cost with accuracy that matches or exceeds a single large model on many tasks. The authors also argue that evaluation is fragmented—MixInstruct, RouterBench, and RouterEval measure different things—and propose the Inference Efficiency Score, a single cost-normalized metric, as the missing common yardstick. In their view, the remaining barriers to wide deployment are multimodality, scalability, energy accounting, and adaptivity, not the basic feasibility of the approach.

What carries the argument

The central object is the Inference Efficiency Score, $\text{IES}(q) = \frac{\alpha Q(q) + (1-\alpha) R(q)}{C(M_k)}$, a single number that divides a utility blend by the full cost of choosing model $M_k$ for query $q$. The utility blend weights a task-quality score $Q(q)$ against a responsiveness score $R(q)$ using a tunable parameter $\alpha$, while the denominator $C(M_k)$ is defined as the sum of seven cost components: compute FLOPs, memory footprint, energy use, latency, financial API cost, scalability load, and a modality-compatibility penalty. Each component carries a coefficient ($\beta_k, \gamma_k, \delta_k, \lambda_k, \mu_k, \rho_k, \sigma_k$) in the paper's formulation, but the paper assigns no values to those coefficients. The score is what converts the survey's qualitative cost–performance comparisons into a quantitative, cost-normalized basis for evaluating both single-shot routing and cumulative cascade decisions; for cascades, the authors suggest a cumulative variant that aggregates the score over the sequence of models visited during escalation.

What would settle it

Take two routing systems, A and B, evaluate both on the same benchmark (for example RouterBench), and compute IES under two defensible coefficient settings: one that weights financial cost heavily (large $\mu$) and one that weights energy heavily (large $\delta$). If the two settings rank A above B in one case and B above A in the other, then IES is not a stable basis for comparing systems, and the paper's main proposed instrument fails its purpose. The paper presents no empirical IES computation, so this test remains open.

Watch

Extended reading notes

Core claim

The paper's claim, stated on its own terms, is that routing and HI frameworks substantially reduce resource consumption without significantly compromising performance, making them well-suited to budget-aware, latency-sensitive, and edge deployments. The authors support this by reviewing fourteen techniques: routing methods such as ZOOTER, MetaLLM, RouteLLM, and Tryage that select one model per query, and cascades such as FrugalGPT, EcoAssistant, Cache & Distil, and Automix that escalate through a model sequence with confidence thresholds. They then systematize the design space with a taxonomy keyed to seven resource constraints (compute, memory, energy, latency, financial cost, scalability, modality) and introduce the Inference Efficiency Score, $\text{IES}(q) = \frac{\alpha Q(q) + (1-\alpha) R(q)}{C(M_k)}$, where $C(M_k)$ is an additive cost over those constraints. The metric is proposed as a unified evaluation basis for future benchmarks; the paper does not itself compute it.

Load-bearing premise

The load-bearing premise is that seven incommensurable quantities—compute, memory, energy, latency, financial cost, scalability load, and modality incompatibility—can be added into one meaningful cost number $C(M_k)$, and that dividing utility by that sum yields a comparable score; the paper gives no values for the coefficients that would make the addition concrete.

Editorial extensions

If this is right

  • Deployments that adopt routing or HI can plausibly expect the savings the reviewed systems report—50–70% for bandit-style routers like MetaLLM, over 50% for EcoAssistant, and up to 98% for FrugalGPT—while keeping accuracy within a few points of the strongest model on most queries.
  • The choice between routing and HI should be driven by which resource is scarce: routing methods are stronger on latency and scalability, while cascades are stronger on financial and compute cost.
  • If IES is adopted, benchmark results from different systems become directly comparable, and future benchmarks can report one normalized score instead of separate accuracy, latency, and cost numbers.
  • The largest gaps identified by the survey are modality support (no reviewed method routes across multimodal LLMs), energy accounting, scalability under bursty load, and adaptive policies that learn from system feedback.
  • Learning-based adaptive routers, such as RL-assisted and bandit formulations, are the most promising near-term direction and can improve efficiency without retraining the underlying LLMs.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the cost coefficients in $C(M_k)$ are never assigned, IES is deployment-relative: one deployment might want to weight financial cost heavily and another energy, so the same two systems could swap order under different weightings; a fixed global 'most efficient system' ranking may not exist.
  • A concrete testable extension is to implement the modality penalty term $C_{\text{modality}}(M_k)$ as a hard constraint in a router: a text-only small model should be skipped whenever the query contains an image or audio component, even if its text-complexity score is high.
  • The survey's own review suggests that feedback-aware routing—using queue length, GPU load, and network latency in the routing decision—should outperform static routers under bursty workloads, but no reviewed system yet demonstrates this.
  • The large savings figures (50–98%) are inherited from the reviewed papers rather than verified head-to-head; a single benchmark that runs all fourteen systems under one cost model and one accuracy metric would be the natural experiment to confirm the survey's central claim.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This survey reviews two families of techniques for efficient multi-LLM inference: routing, which selects one model per query on the basis of predicted performance and cost, and hierarchical inference (HI)/cascading, which escalates a query through a sequence of models until a confidence threshold is met. The paper proposes a deployment-aware taxonomy organized around resource constraints (compute, memory, energy, latency, financial cost, scalability, modality), surveys representative systems (Tryage, ZOOTER, FORC, Routoo, HybridLLM, OptLLM, MetaLLM, RouteLLM, FrugalGPT, EcoAssistant, Cache & Distil, Automix, Efficient Hybrid Decoding, Uncertainty-Based Selection), compares three benchmarks (MixInstruct, ROUTERBENCH, RouterEval), and introduces a unified evaluation metric, the Inference Efficiency Score (IES), defined as a utility-cost ratio. The survey concludes that routing and HI techniques can substantially reduce resource consumption without significant performance loss, and identifies open challenges including multimodality, scalability, evaluation/debugging, adaptive routing, and privacy. The paper's main contribution is presented as a taxonomy plus the IES proposal, with the survey's quantitative claims drawn largely from the cited primary literature.

Significance. If the survey's characterization of the field is accurate, it provides a useful map of a rapidly growing area, and the proposed IES metric could, if made operational, contribute to standardizing comparisons between routing and HI systems. The paper has clear strengths: the description of prior methods aligns well with the cited sources, the tabular comparison across resource constraints (Table II) and the advantages/disadvantages summary (Table III) are useful syntheses, and the benchmark comparison (Table IV) helps clarify what existing evaluation tools do and do not measure. The paper also explicitly situates itself relative to prior surveys and identifies genuinely underexplored directions, such as modality-aware routing and privacy-aware escalation. However, the paper's own novel contribution, the IES metric, is presented only as a formula with an undefined cost model and is never instantiated on any system; the central conclusion's quantitative claims (98% cost savings, 50-70% cost reduction, 75% cost reduction, 10% accuracy gain) are imported from prior work and are not re-derived or validated by the proposed framework.

major comments (4)
  1. [Section III.A and Section IV.C, Eq. (1)] The cost model C(M_k) sums terms with incompatible units: FLOPs, gigabytes of memory, energy, seconds, dollars, request counts, and a unitless modality indicator, each multiplied by unassigned coefficients beta_k, gamma_k, delta_k, lambda_k, mu_k, rho_k, sigma_k. As written, different readers cannot compute the same C(M_k), and the IES in Eq. (1) inherits this problem. The paper never assigns values, specifies normalization, or provides a worked example. Since IES is advertised as a main contribution (Introduction, contribution (ii)), this is a load-bearing gap: either the cost model needs a concrete instantiation with defined units and coefficient values, or the paper should explicitly reframe IES as a conceptual framework requiring deployment-specific calibration, and avoid claiming it as a ready-to-use metric.
  2. [Section IV.A and Table I] The paper is internally inconsistent about the classification of FORC. Section III.B (item 3) and Table I list FORC as a routing technique, while Section IV.A states 'Conversely, HI techniques such as FORC and FrugalGPT focus on minimizing economic costs through selective offloading,' which places FORC in the HI/cascading category. This inconsistency undermines the paper's proposed deployment-aware taxonomy, which is a central contribution. The authors should reconcile the classification, or if FORC's meta-model-based one-shot selection is indeed routing, correct the Section IV.A sentence.
  3. [Section IV.C] The IES is never computed for any system in the paper, and the promised 'cumulative version of IES' for hierarchical systems is mentioned but never defined. Consequently, the paper provides no quantitative support from its own framework for the conclusion that routing and HI 'can substantially reduce resource consumption without significantly compromising performance.' The supporting numbers in Section VI (98% savings, 50-70% reduction, 75% reduction, 10% accuracy gain) come from the cited primary papers. This is acceptable for a survey, but the conclusion should attribute these numbers to the surveyed literature, and the IES proposal should be clearly labeled as a proposal requiring instantiation rather than a validated evaluation tool.
  4. [Section VI] The conclusion's first sentence states the central claim as a demonstrated fact. Given that the paper contributes no new experiments and the IES metric is not applied, the claim should be softened to a synthesis of the surveyed literature, e.g., 'the surveyed systems have reported substantial resource reductions without significant performance loss.' This is a minor wording issue, but it matters because the paper's own metric does not yet provide independent evidence for the claim.
minor comments (5)
  1. [Section III.A, Eq. (2)] The notation M(q) = M1(q), ... is inconsistent: M_k is a model, not a function of q. Use 'M(q) = M1' etc., and similarly in the cascade definition the condition 's1(q) >= tau1' should refer to the score of the currently selected model, not the final output.
  2. [References] Reference [2] for GPT-3 cites Dale's commentary 'GPT-3: What is it good for?' rather than the original GPT-3 paper (Brown et al.), which appears later as [46]; consider citing the primary source at first mention.
  3. [Section I.A] References [27] and [96] are the same work (Varangot-Reille et al., 'Doing More with Less: Implementing Routing Strategies in Large Language Model-based Systems: An Extended Survey'); the duplicate reference number should be cleaned up.
  4. [Section V.A] The phrase 'somehow delaying rigorous comparison of approaches' is informal for a journal; consider replacing 'somehow' with a more precise qualifier such as 'thereby' or 'which has delayed'.
  5. [Section V.D] The discussion of reasoning-capable models and tool-use (paragraph starting 'Routing for Reasoning-Capable Models') is forward-looking and useful, but the claim that cost models 'must be adapted' could benefit from a concrete example of how reasoning depth would be quantified, since the paper elsewhere proposes a specific additive cost model.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the survey's central claims rest on external literature, and the proposed Inference Efficiency Score is an explicit definition rather than a back-derived prediction.

full rationale

The paper is a survey, not a derivation paper. The Section VI claim that routing and HI 'can substantially reduce resource consumption without significantly compromising performance' is supported by numbers imported from the surveyed systems (FrugalGPT up to 98% cost savings, MetaLLM 50-70% reduction, ROUTERBENCH-related 75% cost reduction), which are independent sources outside this paper's own formalism. The only novel formalism, the Inference Efficiency Score (Eq. 1), is introduced by the sentence 'we define a multi-LLM Inference Efficiency Score (IES)...' and is therefore a definition, not a derived consequence; normalizing by the cost model C(M_k) from Section III.A does not make it circular because the paper never claims to predict or fit that cost model. The cost coefficients beta, gamma, delta, lambda, mu, rho, and sigma are never assigned values and no IES value is computed for any benchmark, so the metric is currently uninstantiated and difficult to use, but that is an unverified-contribution problem, not circular reasoning. The self-citations in the paper ([63] and [84]) are not load-bearing: [63] is cited as ongoing research on edge inference feasibility, and [84] is acknowledged only as 'inspiration' for a metric that the paper itself defines; neither citation supplies a contested premise or forbids an alternative. No fitted parameter is renamed as a prediction, and no equation reduces to an earlier equation by construction. Consequently, no circular step is established.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The survey's central contribution, the IES, rests on a cost model with seven unspecified weights and an undefined trade-off parameter alpha; without values or a fitting procedure the metric is not computable. The rest of the paper's assertions inherit the accuracy of the cited literature.

free parameters (3)
  • Cost-scaling coefficients beta_k, gamma_k, delta_k, lambda_k, mu_k, rho_k, sigma_k
    Introduced in Section III.A to combine FLOPs, memory, energy, latency, API cost, load, and modality into a single additive cost; values never assigned, and units are incommensurable.
  • IES weight alpha
    Introduced in Eq. (1) to balance quality Q(q) and responsiveness R(q); no default or fitting procedure is given.
  • Cascade confidence thresholds tau_k
    Used in the HI decision rule in Section III.A; the paper does not specify how they are set.
assumptions (3)
  • ad hoc to paper The additive cost model with heterogeneous units is a valid representation of deployment constraints.
    Section III.A defines C(M_k) as a sum of FLOPs, memory, energy, latency, financial, scalability, and modality terms; this is introduced for this survey and is not derived or empirically grounded.
  • domain assumption Reported results from cited papers (e.g., FrugalGPT's 98% cost savings, EcoAssistant's 10% accuracy gain) are accurate.
    The survey's comparative analysis in Section IV and Table I depends on trusting the original papers' reported numbers.
  • standard math Background transformer and LLM mathematical formulations from Section II are standard.
    Attention mechanism and autoregressive factorization are standard equations treated as known.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Efficient Multi-LLM Inference: Characterization and Analysis of LLM Routing and Hierarchical Techniques." pith.science (2026). https://pith.science/paper/ZDJW2D3G

@misc{pith2026250606579,
  author       = {Pith},
  title        = {Pith review of: Towards Efficient Multi-LLM Inference: Characterization and Analysis of LLM Routing and Hierarchical Techniques},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZDJW2D3G}},
  note         = {Machine review of arXiv:2506.06579}
}
read the original abstract

Recent progress in Language Models (LMs) has dramatically advanced the field of natural language processing (NLP), excelling at tasks like text generation, summarization, and question answering. However, their inference remains computationally expensive and energy intensive, especially in settings with limited hardware, power, or bandwidth. This makes it difficult to deploy LMs in mobile, edge, or cost sensitive environments. To address these challenges, recent approaches have introduced multi LLM intelligent model selection strategies that dynamically allocate computational resources based on query complexity -- using lightweight models for simpler queries and escalating to larger models only when necessary. This survey explores two complementary strategies for efficient LLM inference: (i) routing, which selects the most suitable model based on the query, and (ii) cascading or hierarchical inference (HI), which escalates queries through a sequence of models until a confident response is found. Both approaches aim to reduce computation by using lightweight models for simpler tasks while offloading only when needed. We provide a comparative analysis of these techniques across key performance metrics, discuss benchmarking efforts, and outline open challenges. Finally, we outline future research directions to enable faster response times, adaptive model selection based on task complexity, and scalable deployment across heterogeneous environments, making LLM based systems more efficient and accessible for real world applications.

Figures

Figures reproduced from arXiv: 2506.06579 by the authors.

Figure 1
Figure 1. Efficient LLM inference using different Routing and HI approaches [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Transformer Architecture. the model computes attention scores using the scaled dot￾product attention formula: Attention(Q, K, V) = softmax  QK⊤ √ dk  V. where, Q, K, V ∈ R n×dk are the query, key, and value matrices, respectively. These are obtained by linearly pro￾jecting the input matrix X using trainable weight matrices WQ,WK,WV ∈ R d×dk as follows: Q = XWQ, K = XWK, and V = XWV . The product QK⊤ ∈ R n×n comput… view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. AI Reasoning for Wireless Communications and Networking: A Survey and Perspectives

    cs.NI 2025-09 conditional novelty 4.0 of 10

    A survey that organizes LLM and AI reasoning methods into a taxonomy and maps them onto the physical, link, network, transport, and application layers of wireless networks.

Reference graph

Works this paper leans on

104 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [84]

    Edge-first language model inference: Mod- els, metrics, and tradeoffs,

    S. Jang and R. Morabito, “Edge-first language model inference: Mod- els, metrics, and tradeoffs,” inProceedings of the 2025 IEEE 45th In- ternational Conference on Distributed Computing Systems Workshops (ICDCSW), 2025, to appear

    work page 2025

  2. [1]

    Bert: Pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, “Bert: Pre-training of deep bidirectional transformers for language understanding,”arXiv preprint arXiv:1810.04805, 2018

    arXiv 2018

  3. [2]

    Gpt-3: What is it good for,

    R. Dale, “Gpt-3: What is it good for,”Nat. Lang. Eng, vol. 27, pp. 113–118, 2021

    2021

  4. [3]

    Deepseek llm: Scaling open-source language models with longtermism,

    X. Bi, D. Chen, G. Chen, S. Chen, D. Dai, C. Deng, H. Ding, K. Dong, Q. Du, Z. Fuet al., “Deepseek llm: Scaling open-source language models with longtermism,”arXiv preprint arXiv:2401.02954, 2024

    arXiv 2024

  5. [4]

    Mobile edge intelligence for large language models: A contemporary survey,

    G. Qu, Q. Chen, W. Wei, Z. Lin, X. Chen, and K. Huang, “Mobile edge intelligence for large language models: A contemporary survey,” arXiv preprint arXiv:2407.18921, 2024

    arXiv 2024

  6. [5]

    Faster and lighter llms: A survey on current challenges and way forward,

    A. Chavan, R. Magazine, S. Kushwaha, M. Debbah, and D. Gupta, “Faster and lighter llms: A survey on current challenges and way forward,”arXiv preprint arXiv:2402.01799, 2024

    arXiv 2024

  7. [6]

    A survey of resource-efficient llm and mul- timodal foundation models,

    M. Xu, W. Yin, D. Cai, R. Yi, D. Xu, Q. Wang, B. Wu, Y . Zhao, C. Yang, S. Wanget al., “A survey of resource-efficient llm and mul- timodal foundation models,”arXiv preprint arXiv:2401.08092, 2024

    arXiv 2024

  8. [7]

    Efficient compressing and tuning methods for large language models: A systematic literature review,

    G. I. Kim, S. Hwang, and B. Jang, “Efficient compressing and tuning methods for large language models: A systematic literature review,” ACM Computing Surveys, vol. 57, no. 10, pp. 1–39, 2025

    2025

Show all 104 references
  1. [8]

    Llamafactory: Unified efficient fine-tuning of 100+ language models,

    Y . Zheng, R. Zhang, J. Zhang, Y . Ye, Z. Luo, Z. Feng, and Y . Ma, “Llamafactory: Unified efficient fine-tuning of 100+ language models,” arXiv preprint arXiv:2403.13372, 2024

    2024 arXiv

  2. [9]

    Evaluation of the phi-3-mini slm for identification of texts related to medicine, health, and sports injuries,

    C. Brogly, S. Rjaibi, C. Liang, E. Lam, E. Wang, A. Levitan, S. Paleczny, and M. Cusimano, “Evaluation of the phi-3-mini slm for identification of texts related to medicine, health, and sports injuries,” arXiv preprint arXiv:2504.08764, 2025

    2025

  3. [10]

    Harnessing moderate- sized language models for reliable patient data deidentification in emergency department records: Algorithm development, validation, and implementation study,

    O. Dor ´emus, D. Russon, B. Contrand, A. Guerra-Adames, M. Avalos- Fernandez, C. Gil-Jardin ´e, E. Lagardeet al., “Harnessing moderate- sized language models for reliable patient data deidentification in emergency department records: Algorithm development, validation, and impl...

    2025

  4. [11]

    Overview of small language models in practice,

    R. O. Popov, N. V . Karpenko, and V . V . Gerasimov, “Overview of small language models in practice,” inCEUR Workshop Proceedings, 2025, pp. 164–182

    2025

  5. [12]

    Metallm: A high-performant and cost- efficient dynamic framework for wrapping llms,

    Q. H. Nguyen, D. C. Hoang, J. Decugis, S. Manchanda, N. V . Chawla, and K. D. Doan, “Metallm: A high-performant and cost- efficient dynamic framework for wrapping llms,”arXiv preprint arXiv:2407.10834, 2024

    2024 arXiv

  6. [13]

    Ecoassistant: Using llm assistant more affordably and accurately,

    J. Zhang, R. Krishna, A. H. Awadallah, and C. Wang, “Ecoassistant: Using llm assistant more affordably and accurately,”arXiv preprint arXiv:2310.03046, 2023

    2023 arXiv

  7. [14]

    Frugalgpt: How to use large language models while reducing cost and improving performance,

    L. Chen, M. Zaharia, and J. Zou, “Frugalgpt: How to use large language models while reducing cost and improving performance,” arXiv preprint arXiv:2305.05176, 2023

    2023 arXiv

  8. [15]

    Harnessing the power of multiple minds: Lessons learned from llm routing,

    K. Srivatsa, K. K. Maurya, and E. Kochmar, “Harnessing the power of multiple minds: Lessons learned from llm routing,”arXiv preprint arXiv:2405.00467, 2024

    2024 arXiv

  9. [16]

    Routing to the expert: Efficient reward-guided ensemble of large language models,

    K. Lu, H. Yuan, R. Lin, J. Lin, Z. Yuan, C. Zhou, and J. Zhou, “Routing to the expert: Efficient reward-guided ensemble of large language models,”arXiv preprint arXiv:2311.08692, 2023

    2023 arXiv

  10. [17]

    Routereval: A comprehensive benchmark for routing llms to explore model-level scaling up in llms,

    Z. Huang, G. Ling, V . S. Liang, Y . Lin, Y . Chen, S. Zhong, H. Wu, and L. Lin, “Routereval: A comprehensive benchmark for routing llms to explore model-level scaling up in llms,”arXiv preprint arXiv:2503.10657, 2025

    2025 arXiv

  11. [18]

    Routerbench: A benchmark for multi-llm routing system,

    Q. J. Hu, J. Bieker, X. Li, N. Jiang, B. Keigwin, G. Ranganath, K. Keutzer, and S. K. Upadhyay, “Routerbench: A benchmark for multi-llm routing system,”arXiv preprint arXiv:2403.12031, 2024

    2024 arXiv

  12. [19]

    A survey on efficient inference for large language models,

    Z. Zhou, X. Ning, K. Hong, T. Fu, J. Xu, S. Li, Y . Lou, L. Wang, Z. Yuan, X. Liet al., “A survey on efficient inference for large language models,”arXiv preprint arXiv:2404.14294, 2024

    2024 arXiv

  13. [20]

    Beyond efficiency: A systematic survey of resource-efficient large language models,

    G. Bai, Z. Chai, C. Ling, S. Wang, J. Lu, N. Zhang, T. Shi, Z. Yu, M. Zhu, Y . Zhanget al., “Beyond efficiency: A systematic survey of resource-efficient large language models,”arXiv preprint arXiv:2401.00625, 2024

    2024 arXiv

  14. [21]

    To- wards greener llms: Bringing energy-efficiency to the forefront of llm inference,

    J. Stojkovic, E. Choukse, C. Zhang, I. Goiri, and J. Torrellas, “To- wards greener llms: Bringing energy-efficiency to the forefront of llm inference,”arXiv preprint arXiv:2403.20306, 2024

    2024 arXiv

  15. [22]

    Llm inference serv- ing: Survey of recent advances and opportunities,

    B. Li, Y . Jiang, V . Gadepally, and D. Tiwari, “Llm inference serv- ing: Survey of recent advances and opportunities,”arXiv preprint arXiv:2407.12391, 2024

    2024 arXiv

  16. [23]

    Towards efficient generative large language model serving: A survey from algorithms to systems,

    X. Miao, G. Oliaro, Z. Zhang, X. Cheng, H. Jin, T. Chen, and Z. Jia, “Towards efficient generative large language model serving: A survey from algorithms to systems,”arXiv preprint arXiv:2312.15234, 2023

    2023 arXiv

  17. [24]

    Llm inference unveiled: Survey and roofline model insights,

    Z. Yuan, Y . Shang, Y . Zhou, Z. Dong, Z. Zhou, C. Xue, B. Wu, Z. Li, Q. Gu, Y . J. Leeet al., “Llm inference unveiled: Survey and roofline model insights,”arXiv preprint arXiv:2402.16363, 2024

    2024 arXiv

  18. [25]

    The efficiency spectrum of large language models: An algorithmic survey,

    T. Ding, T. Chen, H. Zhu, J. Jiang, Y . Zhong, J. Zhou, G. Wang, Z. Zhu, I. Zharkov, and L. Liang, “The efficiency spectrum of large language models: An algorithmic survey,”arXiv preprint arXiv:2312.00678, 2023

    2023 arXiv

  19. [26]

    Harnessing multiple large language models: A survey on llm ensemble,

    Z. Chen, J. Li, P. Chen, Z. Li, K. Sun, Y . Luo, Q. Mao, D. Yang, H. Sun, and P. S. Yu, “Harnessing multiple large language models: A survey on llm ensemble,”arXiv preprint arXiv:2502.18036, 2025

    2025 arXiv

  20. [28]

    A survey on collaborative mech- anisms between large and small language models,

    Y . Chen, J. Zhao, and H. Han, “A survey on collaborative mech- anisms between large and small language models,”arXiv preprint arXiv:2505.07460, 2025

    2025 arXiv

  21. [29]

    A survey on mixture of experts,

    W. Cai, J. Jiang, F. Wang, J. Tang, S. Kim, and J. Huang, “A survey on mixture of experts,”arXiv preprint arXiv:2407.06204, 2024

    2024 arXiv

  22. [30]

    Attention is all you need,

    A. Vaswani, “Attention is all you need,”Advances in Neural Informa- tion Processing Systems, 2017

    2017

  23. [31]

    Improving text classification with transformer,

    G. Soyalp, A. Alar, K. Ozkanli, and B. Yildiz, “Improving text classification with transformer,” in2021 6th International Conference on Computer Science and Engineering (UBMK). IEEE, 2021, pp. 707–712

    2021

  24. [32]

    Scaling neural machine translation,

    M. Ott, S. Edunov, D. Grangier, and M. Auli, “Scaling neural machine translation,”arXiv preprint arXiv:1806.00187, 2018

    2018 arXiv

  25. [33]

    Block-skim: Efficient question answering for transformer,

    Y . Guan, Z. Li, Z. Lin, Y . Zhu, J. Leng, and M. Guo, “Block-skim: Efficient question answering for transformer,” inProceedings of the AAAI conference on artificial intelligence, vol. 36, no. 10, 2022, pp. 10 710–10 719

    2022

  26. [34]

    An attentive survey of attention models,

    S. Chaudhari, V . Mithal, G. Polatkan, and R. Ramanath, “An attentive survey of attention models,”ACM Transactions on Intelligent Systems and Technology (TIST), vol. 12, no. 5, pp. 1–32, 2021. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 15

    2021

  27. [35]

    Attention, please! a survey of neural attention models in deep learning,

    A. de Santana Correia and E. L. Colombini, “Attention, please! a survey of neural attention models in deep learning,”Artificial Intelligence Review, vol. 55, no. 8, pp. 6037–6124, 2022

    2022

  28. [36]

    Show and tell: A neural image caption generator,

    O. Vinyals, A. Toshev, S. Bengio, and D. Erhan, “Show and tell: A neural image caption generator,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 3156–3164

    2015

  29. [37]

    Deep learning,

    I. Goodfellow, “Deep learning,” 2016

    2016

  30. [38]

    Deep learning,

    Y . LeCun, Y . Bengio, and G. Hinton, “Deep learning,”nature, vol. 521, no. 7553, pp. 436–444, 2015

    2015

  31. [39]

    Long short-term memory,

    A. Graves and A. Graves, “Long short-term memory,”Supervised sequence labelling with recurrent neural networks, pp. 37–45, 2012

    2012

  32. [40]

    Towards expert- level medical question answering with large language models,

    K. Singhal, T. Tu, J. Gottweis, R. Sayres, E. Wulczyn, L. Hou, K. Clark, S. Pfohl, H. Cole-Lewis, D. Nealet al., “Towards expert- level medical question answering with large language models,”arXiv preprint arXiv:2305.09617, 2023

    2023 arXiv

  33. [41]

    Training data-efficient image transformers & distillation through attention,

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. J ´egou, “Training data-efficient image transformers & distillation through attention,” inInternational conference on machine learning. PMLR, 2021, pp. 10 347–10 357

    2021

  34. [42]

    Splitlora: A split parameter-efficient fine-tuning framework for large language models,

    Z. Lin, X. Hu, Y . Zhang, Z. Chen, Z. Fang, X. Chen, A. Li, P. Vepakomma, and Y . Gao, “Splitlora: A split parameter-efficient fine-tuning framework for large language models,”arXiv preprint arXiv:2407.00952, 2024

    2024 arXiv

  35. [43]

    Albert: A lite bert for self-supervised learning of language representations,

    Z. Lan, “Albert: A lite bert for self-supervised learning of language representations,”arXiv preprint arXiv:1909.11942, 2019

    1909 arXiv

  36. [44]

    Exploring the limits of transfer learning with a unified text-to-text transformer,

    C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y . Zhou, W. Li, and P. J. Liu, “Exploring the limits of transfer learning with a unified text-to-text transformer,”Journal of machine learning research, vol. 21, no. 140, pp. 1–67, 2020

    2020

  37. [45]

    Gpt-4 is here: what scientists think,

    K. Sanderson, “Gpt-4 is here: what scientists think,”Nature, vol. 615, no. 7954, p. 773, 2023

    2023

  38. [46]

    Language mod- els are few-shot learners,

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askellet al., “Language mod- els are few-shot learners,”Advances in neural information processing systems, vol. 33, pp. 1877–1901, 2020

    1901

  39. [47]

    Multimodal deep learning

    J. Ngiam, A. Khosla, M. Kim, J. Nam, H. Lee, A. Y . Nget al., “Multimodal deep learning.” inICML, vol. 11, 2011, pp. 689–696

    2011

  40. [48]

    Multimodal machine learning: A survey and taxonomy,

    T. Baltru ˇsaitis, C. Ahuja, and L.-P. Morency, “Multimodal machine learning: A survey and taxonomy,”IEEE transactions on pattern analysis and machine intelligence, vol. 41, no. 2, pp. 423–443, 2018

    2018

  41. [49]

    Multimodal transformer for unaligned multimodal language sequences,

    Y .-H. H. Tsai, S. Bai, P. P. Liang, J. Z. Kolter, L.-P. Morency, and R. Salakhutdinov, “Multimodal transformer for unaligned multimodal language sequences,” inProceedings of the conference. Association for computational linguistics. Meeting, vol. 2019. NIH Public Access, 2019...

    2019

  42. [50]

    Gpt-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkatet al., “Gpt-4 technical report,”arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  43. [51]

    Crossner: Evaluating cross-domain named entity recognition,

    Z. Liu, Y . Xu, T. Yu, W. Dai, Z. Ji, S. Cahyawijaya, A. Madotto, and P. Fung, “Crossner: Evaluating cross-domain named entity recognition,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 15, 2021, pp. 13 452–13 460

    2021

  44. [52]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning. PMLR, 2021, pp. 8748–8763

    2021

  45. [53]

    On the opportunities and risks of foundation models,

    R. Bommasani, D. A. Hudson, E. Adeli, R. Altman, S. Arora, S. von Arx, M. S. Bernstein, J. Bohg, A. Bosselut, E. Brunskillet al., “On the opportunities and risks of foundation models,”arXiv preprint arXiv:2108.07258, 2021

    2021 arXiv

  46. [54]

    Xtext language engineering for everyone,

    E. Foundation, “Xtext language engineering for everyone,” https:// eclipse.dev/Xtext/, 2023, accessed: date-of-access

    2023

  47. [55]

    Align before fuse: Vision and language representation learning with momentum distillation,

    J. Li, R. Selvaraju, A. Gotmare, S. Joty, C. Xiong, and S. C. H. Hoi, “Align before fuse: Vision and language representation learning with momentum distillation,”Advances in neural information processing systems, vol. 34, pp. 9694–9705, 2021

    2021

  48. [56]

    Training language models to follow instructions with human feedback,

    L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. Wainwright, P. Mishkin, C. Zhang, S. Agarwal, K. Slama, A. Rayet al., “Training language models to follow instructions with human feedback,”Advances in neural information processing systems, vol. 35, pp. 27 730–27 744, 2022

    2022

  49. [57]

    Hymba: A hybrid-head architecture for small language models,

    X. Dong, Y . Fu, S. Diao, W. Byeon, Z. Chen, A. S. Mahabaleshwarkar, S.-Y . Liu, M. Van Keirsbilck, M.-H. Chen, Y . Suharaet al., “Hymba: A hybrid-head architecture for small language models,”arXiv preprint arXiv:2411.13676, 2024

    2024 arXiv

  50. [58]

    A comprehensive survey of small language models in the era of large language models: Techniques, en- hancements, applications, collaboration with llms, and trustworthiness,

    F. Wang, Z. Zhang, X. Zhang, Z. Wu, T. Mo, Q. Lu, W. Wang, R. Li, J. Xu, X. Tanget al., “A comprehensive survey of small language models in the era of large language models: Techniques, en- hancements, applications, collaboration with llms, and trustworthiness,” arXiv preprint...

    2024 arXiv

  51. [59]

    Small language models: Survey, measurements, and insights,

    Z. Lu, X. Li, D. Cai, R. Yi, F. Liu, X. Zhang, N. D. Lane, and M. Xu, “Small language models: Survey, measurements, and insights,”arXiv preprint arXiv:2409.15790, 2024

    2024 arXiv

  52. [60]

    Hallucination is inevitable: An innate limitation of large language models,

    Z. Xu, S. Jain, and M. Kankanhalli, “Hallucination is inevitable: An innate limitation of large language models,”arXiv preprint arXiv:2401.11817, 2024

    2024 arXiv

  53. [61]

    A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,

    L. Huang, W. Yu, W. Ma, W. Zhong, Z. Feng, H. Wang, Q. Chen, W. Peng, X. Feng, B. Qinet al., “A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,”ACM Transactions on Information Systems, 2024

    2024

  54. [62]

    Towards trustworthy llms: a review on debiasing and dehallucinating in large language models,

    Z. Lin, S. Guan, W. Zhang, H. Zhang, Y . Li, and H. Zhang, “Towards trustworthy llms: a review on debiasing and dehallucinating in large language models,”Artificial Intelligence Review, vol. 57, no. 9, p. 243, 2024

    2024

  55. [63]

    Sometimes painful but certainly promising: Feasibility and trade-offs of language model inference at the edge,

    M. Abstreiter, S. Tarkoma, and R. Morabito, “Sometimes painful but certainly promising: Feasibility and trade-offs of language model inference at the edge,”arXiv preprint arXiv:2503.09114, 2025

    2025

  56. [64]

    Tryage: Real-time, intelligent rout- ing of user prompts to large language model,

    S. N. Hari and M. Thomson, “Tryage: Real-time, intelligent rout- ing of user prompts to large language model,”arXiv preprint arXiv:2308.11601, 2023

    2023 arXiv

  57. [65]

    Fly-swat or cannon? cost- effective language model choice via meta-modeling,

    M. ˇSakota, M. Peyrard, and R. West, “Fly-swat or cannon? cost- effective language model choice via meta-modeling,” inProceedings of the 17th ACM International Conference on Web Search and Data Mining, 2024, pp. 606–615

    2024

  58. [66]

    Leeroo orchestrator: Elevating llms performance through model integration,

    A. Mohammadshahi, A. Shaikh, and M. Yazdani, “Leeroo orchestrator: Elevating llms performance through model integration,”arXiv preprint arXiv:2401.13979, 2024

    2024 arXiv

  59. [67]

    Hybrid llm: Cost-efficient and quality-aware query routing,

    D. Ding, A. Mallick, C. Wang, R. Sim, S. Mukherjee, V . Ruhle, L. V . Lakshmanan, and A. H. Awadallah, “Hybrid llm: Cost-efficient and quality-aware query routing,”arXiv preprint arXiv:2404.14618, 2024

    2024 arXiv

  60. [68]

    Optllm: Optimal assignment of queries to large language models,

    Y . Liu, H. Zhang, Y . Miao, V .-H. Le, and Z. Li, “Optllm: Optimal assignment of queries to large language models,”arXiv preprint arXiv:2405.15130, 2024

    2024 arXiv

  61. [69]

    Chatgpt,

    OpenAI and Microsoft, “Chatgpt,” https://chatgpt.com/, 2022, ac- cessed: date-of-access

    2022

  62. [70]

    Amazon titan in amazon bedrock,

    Amazon, “Amazon titan in amazon bedrock,” https://aws.amazon.com/ bedrock/amazon-models/, 2023, accessed: date-of-access

    2023

  63. [71]

    Anthropic, “Claude,” https://claude.ai/, 2023, accessed: date-of-access

    2023

  64. [72]

    MetaAI, “Llama,” https://www.llama.com/, 2023, accessed: date-of- access

    2023

  65. [73]

    Routellm: Learning to route llms with preference data,

    I. Ong, A. Almahairi, V . Wu, W.-L. Chiang, T. Wu, J. E. Gonzalez, M. W. Kadous, and I. Stoica, “Routellm: Learning to route llms with preference data,”arXiv preprint arXiv:2406.18665, 2024

    2024 arXiv

  66. [74]

    Cache & dis- til: Optimising api calls to large language models,

    G. Ram ´ırez, M. Lindemann, A. Birch, and I. Titov, “Cache & dis- til: Optimising api calls to large language models,”arXiv preprint arXiv:2310.13561, 2023

    2023 arXiv

  67. [75]

    Automix: Automatically mixing language models,

    P. Aggarwal, A. Madaan, A. Anand, S. P. Potharaju, S. Mishra, P. Zhou, A. Gupta, D. Rajagopal, K. Kappaganthu, Y . Yanget al., “Automix: Automatically mixing language models,”arXiv preprint arXiv:2310.12963, 2023

    2023 arXiv

  68. [76]

    Efficient hybrid inference for llms: Reward- based token modelling with selective cloud assistance,

    A. MS, J. VG, and D. PS, “Efficient hybrid inference for llms: Reward- based token modelling with selective cloud assistance,”arXiv preprint arXiv:2409.13757, 2024

    2024 arXiv

  69. [77]

    Optimising calls to large lan- guage models with uncertainty-based two-tier selection,

    G. Ram ´ırez, A. Birch, and I. Titov, “Optimising calls to large lan- guage models with uncertainty-based two-tier selection,”arXiv preprint arXiv:2405.02134, 2024

    2024 arXiv

  70. [78]

    Query by committee,

    H. S. Seung, M. Opper, and H. Sompolinsky, “Query by committee,” in Proceedings of the fifth annual workshop on Computational learning theory, 1992, pp. 287–294

    1992

  71. [79]

    A review on edge large language models: Design, execution, and applications,

    Y . Zheng, Y . Chen, B. Qian, X. Shi, Y . Shu, and J. Chen, “A review on edge large language models: Design, execution, and applications,” ACM Computing Surveys, vol. 57, no. 8, pp. 1–35, 2025

    2025

  72. [80]

    On-device language models: A comprehensive review,

    J. Xu, Z. Li, W. Chen, Q. Wang, X. Gao, Q. Cai, and Z. Ling, “On-device language models: A comprehensive review,”arXiv preprint arXiv:2409.00088, 2024

    2024 arXiv

  73. [81]

    Llm-blender: Ensembling large language models with pairwise ranking and generative fusion,

    D. Jiang, X. Ren, and B. Y . Lin, “Llm-blender: Ensembling large language models with pairwise ranking and generative fusion,”arXiv preprint arXiv:2306.02561, 2023

    2023 arXiv

  74. [82]

    Research

    I. Research. (2024) Llm routers: Efficient routing of large language models. Accessed: 2025-06-06. [Online]. Available: https://research. ibm.com/blog/LLM-routers?utm source=chatgpt.com JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 16

    2024

  75. [83]

    P. AI. (2024) Llm routing: Ai costs optimization without sacrificing quality. Accessed: 2025-06-06. [Online]. Available: https://blog.premai. io/llm-routing-ai-costs-optimisation-without-sacrificing-quality/?utm source=chatgpt.com

    2024

  76. [85]

    Ultraeval: A lightweight platform for flexible and comprehensive evaluation for llms,

    C. He, R. Luo, S. Hu, Y . Zhao, J. Zhou, H. Wu, J. Zhang, X. Han, Z. Liu, and M. Sun, “Ultraeval: A lightweight platform for flexible and comprehensive evaluation for llms,”arXiv preprint arXiv:2404.07584, 2024

    2024 arXiv

  77. [86]

    Etalon: Holistic performance evaluation framework for llm inference systems,

    A. Agrawal, A. Agarwal, N. Kedia, J. Mohan, S. Kundu, N. Kwa- tra, R. Ramjee, and A. Tumanov, “Etalon: Holistic performance evaluation framework for llm inference systems,”arXiv preprint arXiv:2407.07000, 2024

    2024 arXiv

  78. [87]

    Mm-llms: Recent advances in multimodal large language models,

    D. Zhang, Y . Yu, J. Dong, C. Li, D. Su, C. Chu, and D. Yu, “Mm-llms: Recent advances in multimodal large language models,”arXiv preprint arXiv:2401.13601, 2024

    2024 arXiv

  79. [88]

    Mquant: Unleashing the inference potential of multimodal large language models via full static quantization,

    J. Yu, S. Zhou, D. Yang, S. Wang, S. Li, X. Hu, C. Xu, Z. Xu, C. Shu, and Z. Yuan, “Mquant: Unleashing the inference potential of multimodal large language models via full static quantization,”arXiv preprint arXiv:2502.00425, 2025

    2025 arXiv

  80. [89]

    Multimodal large language models in health care: applications, challenges, and future outlook,

    R. AlSaad, A. Abd-Alrazaq, S. Boughorbel, A. Ahmed, M.-A. Renault, R. Damseh, and J. Sheikh, “Multimodal large language models in health care: applications, challenges, and future outlook,”Journal of medical Internet research, vol. 26, p. e59505, 2024

    2024

  81. [90]

    Ma-lmm: Memory-augmented large multimodal model for long-term video understanding,

    B. He, H. Li, Y . K. Jang, M. Jia, X. Cao, A. Shah, A. Shrivastava, and S.-N. Lim, “Ma-lmm: Memory-augmented large multimodal model for long-term video understanding,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 13 504–13 514

    2024

  82. [91]

    Lexical complexity predic- tion: An overview,

    K. North, M. Zampieri, and M. Shardlow, “Lexical complexity predic- tion: An overview,”ACM Computing Surveys, vol. 55, no. 9, pp. 1–42, 2023

    2023

  83. [92]

    Collaborative cross- modal fusion with large language model for recommendation,

    Z. Liu, H. Zhang, K. Dong, and Y . Fang, “Collaborative cross- modal fusion with large language model for recommendation,” in Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, 2024, pp. 1565–1574

    2024

  84. [93]

    Communication-efficient distributed on-device llm inference over wireless networks,

    K. Zhang, H. He, S. Song, J. Zhang, and K. B. Letaief, “Communication-efficient distributed on-device llm inference over wireless networks,”arXiv preprint arXiv:2503.14882, 2025

    2025 arXiv

  85. [94]

    Intelligent router for llm workloads: Improving performance through workload-aware scheduling,

    K. Jain, A. Parayil, A. Mallick, E. Choukse, X. Qin, J. Zhang, ´I. Goiri, R. Wang, C. Bansal, V . R ¨uhleet al., “Intelligent router for llm workloads: Improving performance through workload-aware scheduling,”arXiv preprint arXiv:2408.13510, 2024

    2024 arXiv

  86. [95]

    Llm cas- cade with multi-objective optimal consideration,

    K. Zhang, L. Peng, C. Wang, A. Go, and X. Liu, “Llm cas- cade with multi-objective optimal consideration,”arXiv preprint arXiv:2410.08014, 2024

    2024 arXiv

  87. [96]

    Doing more with less–implementing routing strategies in large language model-based systems: An extended survey,

    C. Varangot-Reille, C. Bouvard, A. Gourru, M. Ciancone, M. Scha- effer, and F. Jacquenet, “Doing more with less–implementing routing strategies in large language model-based systems: An extended survey,” arXiv preprint arXiv:2502.00409, 2025

    2025 arXiv

  88. [97]

    Xai meets llms: A survey of the relation between explainable ai and large language models,

    E. Cambria, L. Malandri, F. Mercorio, N. Nobani, and A. Seveso, “Xai meets llms: A survey of the relation between explainable ai and large language models,”arXiv preprint arXiv:2407.15248, 2024

    2024 arXiv

  89. [98]

    Pickllm: Context- aware rl-assisted large language model routing,

    D. Sikeridis, D. Ramdass, and P. Pareek, “Pickllm: Context- aware rl-assisted large language model routing,”arXiv preprint arXiv:2412.12170, 2024

    2024 arXiv

  90. [99]

    Text sanitization beyond specific domains: Zero-shot redaction & substitution with large lan- guage models,

    F. Albanese, D. Ciolek, and N. D’Ippolito, “Text sanitization beyond specific domains: Zero-shot redaction & substitution with large lan- guage models,”arXiv preprint arXiv:2311.10785, 2023

    2023 arXiv

  91. [100]

    Trusted llm inference on the edge with smart contracts,

    R. Karanjai and W. Shi, “Trusted llm inference on the edge with smart contracts,” in2024 IEEE International Conference on Blockchain and Cryptocurrency (ICBC). IEEE, 2024, pp. 1–7

    2024

  92. [101]

    Encryption-friendly llm architecture,

    D. Rho, T. Kim, M. Park, J. W. Kim, H. Chae, E. K. Ryu, and J. H. Cheon, “Encryption-friendly llm architecture,”arXiv preprint arXiv:2410.02486, 2024

    2024 arXiv

  93. [102]

    Anonymizing medical documents with local, privacy preserving large language models: The llm-anonymizer,

    I. C. Wiest, M.-E. Leßmann, F. Wolf, D. Ferber, M. Van Treeck, J. Zhu, M. P. Ebert, C. B. Westphalen, M. Wermke, and J. N. Kather, “Anonymizing medical documents with local, privacy preserving large language models: The llm-anonymizer,”medRxiv, pp. 2024–06, 2024

    2024

  94. [103]

    Llm hallucinations and personal data accuracy: can they really co-exist?

    S. Rossello, “Llm hallucinations and personal data accuracy: can they really co-exist?”European Law Blog (https://www. europeanlawblog. eu/pub/2klfhf06/release/1), 2025

    2025

  95. [104]

    Federatedscope-llm: A comprehensive package for fine-tuning large language models in federated learning,

    W. Kuang, B. Qian, Z. Li, D. Chen, D. Gao, X. Pan, Y . Xie, Y . Li, B. Ding, and J. Zhou, “Federatedscope-llm: A comprehensive package for fine-tuning large language models in federated learning,” inProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data ...

    2024

  96. [105]

    Fellas: Enhancing federated sequential recommendation with llm as external services,

    W. Yuan, C. Yang, G. Ye, T. Chen, N. Q. V . Hung, and H. Yin, “Fellas: Enhancing federated sequential recommendation with llm as external services,”ACM Transactions on Information Systems, 2024

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.