Pith. sign in

REVIEW 3 major objections 6 minor 28 references

Under a fixed per-query budget and imperfect verifier, an online policy that spends each unit on either resampling the committed model or rerouting to another attains a better cost–quality tradeoff than single-commit routers, cascades, or p

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-13 06:31 UTC pith:LQQHYSMK

load-bearing objection Solid systems paper that unifies resampling and rerouting under one per-query budget; the Pareto evidence is real, but the imperfect-verifier ablation still feeds true labels into allocation. the 3 major comments →

arxiv 2607.08665 v2 pith:LQQHYSMK submitted 2026-07-09 cs.LG

Resample or Reroute? Budget-Aware Test-Time Model Selection for Large Language Models

Teng-Ruei Chen This is my paper
classification cs.LG
keywords model routingtest-time computebudget-aware inferencebest-of-K samplingLLM cascadesresample-or-reroutecost-quality Paretoimperfect verifier
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Serving a query with large language models is no longer just which model to call; it is how to spend a limited inference budget. Prior routing work commits once per query, while best-of-K work resamples one fixed model; neither treats those two moves as competing uses of the same per-query dollars. This paper formalizes budget-aware test-time model selection: given a cost limit and an imperfect verifier, decide at each step whether to draw again from a model already tried or to open a new model, so that expected correctness is maximized. The proposed resample-or-reroute policy scores every affordable action by estimated marginal correctness per unit cost and greedily takes the best. Replay on multi-draw correctness tensors from an eleven-model open-weight pool across four benchmarks shows the policy lies on a favorable cost–quality front relative to single-route, one-commit router, budget-aware best-of-K, cascade, and random baselines, with the largest lift on the most heterogeneous set. The advantage shrinks as verifier quality falls, and holds under a realistic price vector and certain deployable verifiers.

Core claim

When resampling a committed model and rerouting to an alternative compete for one per-query cost budget under an imperfect verifier, an online policy that always spends the next unit of budget on the action with highest estimated marginal correctness per unit cost attains a favorable cost–quality Pareto front versus single-route, one-commit-router, budget-aware best-of-K, cascade, and random-allocation baselines on the tested eleven-model pools, with largest gains on the most heterogeneous benchmark and with those gains gated by verifier quality.

What carries the argument

Resample-or-reroute (RoR): a greedy (or UCB) online allocation that maintains a posterior success probability for each model on the current query and at every step chooses the affordable action—one more sample of a used model or the first sample of a new model—with the highest estimated increase in final correctness per unit cost, optionally early-stopping when the verifier accepts.

Load-bearing premise

The policy needs a usable verifier whose signal both guides which draw to take next and selects the final answer; without that, the advantage over simply resampling one strong model shrinks or disappears.

What would settle it

Replay the same multi-draw tensors with a fully noisy guidance signal (so the belief counts themselves are corrupted by the same imperfect verifier used for final selection) and check whether RoR still dominates budget-aware best-of-K and cascade on the heterogeneous GPQA-style pool at matched mid budget.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • On heterogeneous model pools with a trustworthy verifier, systems should treat resampling and rerouting as interchangeable budget spends rather than separate mechanisms.
  • When the answer space is open-ended or a reliable partial checker exists (e.g., unit tests for code), test-time recovery of specialist advantage is available without a perfect oracle.
  • On near-saturated or homogeneous pools the simpler budget-aware best-of-K already captures most of the gain, so full RoR is unnecessary.
  • Agreement-based self-consistency is a usable free verifier for free-form answers but fails on multiple-choice tasks where wrong answers collide easily.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Any serving stack that already has a process-reward model or execution harness can plug that signal into RoR without changing the allocation rule, turning an offline recovery guarantee into a live policy.
  • The same marginal-gain framing could be applied inside a single model’s token budget (continue generating vs. restart or switch decoding strategy), not only across models.
  • If future model catalogs keep growing more specialized rather than uniformly better, the value of the reroute branch will rise relative to pure resampling.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper formulates budget-aware test-time model selection for LLMs: under a fixed per-query cost budget and an imperfect verifier, allocate each unit of budget between resampling a committed model and rerouting to an alternative so as to maximize expected correctness. It proposes an online resample-or-reroute (RoR) policy that greedily spends the next unit of budget on the action with highest estimated marginal correctness per unit cost (with a UCB variant and an oracle-allocation ceiling), and grounds the preference for resampling vs. rerouting in the recoverability asymmetry of a companion analysis. On newly regenerated multi-draw correctness tensors (k=30, seed-aligned) from an eleven-model open-weight pool over GSM8K, MATH-500, GPQA-Diamond, and HumanEval+, replay experiments report that RoR attains a favorable cost–quality Pareto front relative to single-route, one-commit router, budget-aware best-of-K, cascade, and random-allocation baselines, with the largest gains on heterogeneous GPQA (0.892 vs 0.644 best-of-K at matched mid budget). An ablation claims the gains are verifier-gated; robustness checks include a provider price vector, an agreement-based verifier, and a partial-test code verifier.

Significance. The framing of resampling and rerouting as competing uses of one explicit per-query budget under an imperfect verifier is a clean and practically relevant contribution that prior routing, cascade, and best-of-N lines treat separately. The empirical protocol is stronger than average for this area: seed-aligned multi-draw tensors, 50/50 train/test priors, 20 draw orderings, budget sweeps, sensitivity on the prior pseudo-count and train fraction, real-price replay, and two deployable verifiers (agreement; partial code tests). The CPU-only replay from released tensors is a genuine reproducibility strength. If the Pareto and verifier-gated claims hold under a fully noisy guidance signal, the work would give a deployable rule for when to spend budget on more samples of the same model versus switching models, with clear regime dependence (saturated vs. heterogeneous pools).

major comments (3)
  1. Section VI (Verifier paragraph) and the q-ablation in Section VII / Figs. 2–3: the parametric imperfect-verifier study degrades only final selection, while “the per-draw guidance signal (the verified counts w_im in the belief update) still uses the recorded correctness labels.” Consequently Eq. (2) and the marginal-gain rule (Eq. 3 / Algorithm 1) continue to receive perfect w_im even at q=0.6. A real imperfect verifier would corrupt both selection and the online belief that decides resample vs. reroute. The residual GPQA lead at q=0.6 (0.538 vs 0.444 best-of-K) and the central claim that “gains are verifier-gated” are therefore measured under an optimistic allocation oracle. Either re-run the ablation with noisy w_im (or a noisy continuous score feeding the belief), or substantially narrow the claim and treat Tables VIII–IX as the primary imperfect-verifier evidence.
  2. Section IV.B and Algorithm 1: the marginal-gain estimator is fully specified only under early stopping, where \Delta corr(a) reduces to ˆp_im and the rule becomes arg max ˆp_im/c_m. For the imperfect-verifier regime (q<1), early stopping is disabled and final selection is a soft mixture of “any correct” and “fraction correct.” It is not clear how \Delta corr is estimated in that case, nor whether the same greedy ranking remains well-defined when the verifier score is continuous or noisy. Please state the general estimator used in the q<1 sweeps and confirm that the reported fronts use that estimator rather than the early-stopping reduction.
  3. Section VI (Baselines) and Table V: the one-commit router’s cost is “proxied by one draw of the most accurate model” because the tensors store only the routed outcome. That makes the router appear as expensive as single-routing the best model (cost 32.0 on three of four benchmarks) and weakens the comparison to learned routers, which are often cheaper on easy queries. Either regenerate routed model identities so cost can be charged correctly, or present the router only as an accuracy reference at its true (unknown) cost and avoid matched-cost claims against it.
minor comments (6)
  1. Table I is useful but the “imperfect verifier” column for cascades/routers is a bit absolute; many cascades already use confidence gates. A short footnote clarifying the distinction (parametric imperfect verifier + joint resample/reroute budget) would help.
  2. Figure 1 caption and Section VII: the low-budget GPQA exception (single-commit beats RoR below cost ~15) is important; consider marking that region on the plot or calling it out in the abstract’s “favorable Pareto” language so readers do not over-read dominance at all budgets.
  3. Section IV.A: the claim that the policy is “insensitive to s over an order of magnitude” is only partly supported by Table VI (s=0.5 and s=2 match; s=8 costs 11.2 points on GPQA). Soften the wording to “stable for moderate s; large s over-anchors on heterogeneous pools.”
  4. Notation: p_im is introduced as an unknown success probability, but the replay uses finite without-replacement draws from k=30. A sentence on finite-population bias (or that k=30 is treated as the generative process) would avoid a small mismatch between formulation and evaluation.
  5. Related work: the reject-option and cascade citations are appropriate; a brief pointer to recent process-reward / verifier literature for LLMs would better situate the “plug in a concrete verifier” discussion in Section VIII.
  6. Typos / polish: “test-timeresampling” and “budget-aware test-time model selection” spacing in the abstract; “Onsaturated” / “Onintermediate” / “Onhard” word-joining in Section VII; “arXiv:2607.08665v2” header is fine for preprint but should be cleaned for journal production.

Circularity Check

1 steps flagged

Empirical methods paper with ordinary train-split priors and conceptual companion grounding; no derivation reduces to its inputs by construction.

specific steps
  1. self citation load bearing [Section V (Theoretical Grounding); also Abstract and Introduction relation paragraph]
    "The recoverability asymmetry [8] states that the per-instance selection floor is closed by no single-commit router but is recovered by resampling at the oracle's budget. The proposed policy inherits two consequences. First, when a committed model's reproducible success probability is high, resampling has larger marginal correctness-per-cost than rerouting, so the greedy rule prefers resampling—matching the asymmetry."

    The policy's qualitative preference for resampling vs. rerouting is justified by citing the companion paper [8] by the same author. This is conceptual framing, not a self-proving loop: the central claims (Pareto fronts, matched-budget accuracies, verifier-gated gains) are independent empirical measurements on regenerated tensors, not consequences forced by the companion theorem. Mild and non-load-bearing.

full rationale

The paper formulates a budgeted correctness objective (Eq. 1), defines an online posterior mean belief (Eq. 2) from offline train-split priors plus online verified counts, and allocates by estimated marginal correctness per unit cost (Eq. 3 / Algorithm 1). Offline priors are calibrated on a held-out train half and evaluated on the test half; that is standard empirical Bayes, not a fitted identity renamed as prediction. The recoverability asymmetry is cited from the companion analysis only as conceptual grounding for when resampling should beat rerouting (Section V); the reported Pareto fronts, matched-budget numbers, and verifier ablations are independent replay measurements on regenerated multi-draw tensors against external baselines. The imperfect-verifier ablation that holds allocation labels fixed while degrading only final selection is an optimistic experimental design choice (and a transfer risk), not a circular reduction of a claimed derivation. No uniqueness theorem, ansatz, or self-definitional identity forces the central empirical claim. Score 2 reflects only the mild, non-load-bearing self-citation to the companion paper.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 2 invented entities

The central empirical claim rests on a small set of modeling choices rather than new physical entities: a per-query hard budget, Bernoulli-like per-draw correctness probabilities, a parametric or agreement/execution verifier, offline accuracy priors, and a cost model (parameter count or provider prices). The policy itself is an invented control rule, not an unobserved particle. Most background is standard bandit/greedy allocation and LLM serving practice; the load-bearing domain assumptions are verifier usability and that offline multi-draw tensors are faithful enough for allocation comparisons.

free parameters (4)
  • prior pseudo-count s = 2 (default)
    Controls how strongly offline accuracy anchors the online posterior mean; default s=2, with sensitivity shown over {0.5,2,8}. Directly shapes resample-vs-reroute switching.
  • verifier quality q = main results q=1; ablation q in {0.6,0.8,1.0}
    Parametric interpolation between perfect early-stop selection and random pick among draws; used to claim gains are verifier-gated.
  • agreement consensus threshold A = 2
    Label-free verifier stops when an answer cluster reaches A matching draws; chosen as A=2.
  • per-query budget grid B = 12-point grid; mid point ~26 size-proxy units
    Operating points from c_min to 6 c_max (12 points); matched mid-budget comparisons use the point nearest mean cost 26 under the size proxy.
axioms (6)
  • domain assumption Each draw from model m on query i is an independent Bernoulli trial with unknown success probability p_im under fixed decoding settings.
    Problem formulation Section III and multi-draw replay protocol; standard stochastic-decoding assumption.
  • domain assumption A usable verifier exists that can score candidates for both early stopping / final selection and (in the main setup) allocation guidance.
    Stated throughout Sections III–IV and shown load-bearing by the q ablation and agreement collapse on GPQA.
  • domain assumption Per-draw cost can be represented by a scalar c_m (parameter count proxy, or a provider price vector) that is known and additive under a hard per-query budget B.
    Section VI Cost paragraph and real-price replay; ordering matters more than exact dollars for the mechanism.
  • domain assumption Offline train-split accuracies \bar p_m are informative enough priors for online posterior means on held-out queries.
    Belief update Eq. (2); sensitivity table shows coarse ranking suffices.
  • ad hoc to paper Without-replacement replay of seed-aligned precomputed draws is a faithful evaluation of allocation policies for correctness-based rewards.
    Experimental protocol Section VI; authors themselves note it misses prompt-adaptive behaviors (Section VIII).
  • domain assumption Recoverability asymmetry from the companion analysis explains when resampling beats rerouting.
    Section V grounding; imported rather than re-derived.
invented entities (2)
  • Budget-aware test-time model selection problem no independent evidence
    purpose: Formalize resampling and rerouting as competing actions under one per-query cost budget and imperfect verifier.
    Core problem statement in Section III; not previously posed this way according to the paper.
  • Resample-or-Reroute (RoR) allocation policy independent evidence
    purpose: Online greedy/UCB rule maximizing estimated marginal correctness per unit cost, with early stopping.
    Algorithm 1 and Section IV; the deployable method being evaluated.

pith-pipeline@v1.1.0-grok45 · 19640 in / 3836 out tokens · 32898 ms · 2026-07-13T06:31:53.538957+00:00 · methodology

0 comments
read the original abstract

Routing among large language models (LLMs) trades response quality against serving cost, motivated by the reported gap between deployed routers and a per-instance oracle. Recent analysis shows that test-time resampling can recover per-instance selection headroom that no single-commit router captures; however, that guarantee holds only under an idealized oracle equipped with correctness labels and an unconstrained budget, neither of which a deployed system has. To the best of our knowledge, no previous work treats resampling the committed model and rerouting to an alternative model as competing uses of a single per-query cost budget. Therefore, this work formulates budget-aware test-time model selection: given a per-query budget and an imperfect verifier, allocate each unit of budget between resampling and rerouting so that expected correctness is maximized. An online resample-or-reroute (RoR) allocation policy driven by estimated marginal correctness per unit cost is proposed, and its behavior is grounded in the recoverability asymmetry between selection and sampling. Replay experiments on newly regenerated multi-draw correctness tensors from an eleven-model open-weight pool over four benchmarks of differing difficulty show that the proposed RoR policy attains a favorable cost-quality Pareto front relative to single-route, one-commit-router, budget-aware best-of-K, cascade, and random-allocation baselines for the tested pools, with the largest gains on the most heterogeneous benchmark; an ablation further shows the gains are verifier-gated, shrinking as verifier quality degrades, and robustness replays under a provider price vector and a label-free agreement verifier delineate where the conclusions carry over.

Figures

Figures reproduced from arXiv: 2607.08665 by Teng-Ruei Chen.

Figure 1
Figure 1. Figure 1: Cost–quality Pareto fronts on the four regenerated pools (verifier [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Verifier-quality ablation: policy accuracy at matched mid budget as [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Cost–quality fronts of the budget-scalable policies under a degraded verifier ( [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 5 linked inside Pith

  1. [1]

    Measuring and improving the energy efficiency of large language models inference,

    M. F. Argerich and M. Patiño-Martínez, “Measuring and improving the energy efficiency of large language models inference,”IEEE Access, vol. 12, pp. 80 194–80 207, 2024

    2024

  2. [2]

    Ma- chine learning (ML)-centric resource management in cloud computing: A review and future directions,

    T. Khan, W. Tian, G. Zhou, S. Ilager, M. Gong, and R. Buyya, “Ma- chine learning (ML)-centric resource management in cloud computing: A review and future directions,”Journal of Network and Computer Applications, vol. 204, p. 103405, 2022

    2022

  3. [3]

    Survey of different large language model architectures: Trends, benchmarks, and challenges,

    M. Shao, A. Basit, R. Karri, and M. Shafique, “Survey of different large language model architectures: Trends, benchmarks, and challenges,” IEEE Access, vol. 12, pp. 188 664–188 706, 2024

    2024

  4. [4]

    RouterBench: A benchmark for multi-LLM routing system,

    Q. J. Huet al., “RouterBench: A benchmark for multi-LLM routing system,”arXiv preprint arXiv:2403.12031, 2024

    Pith/arXiv arXiv 2024

  5. [5]

    LLMRouterBench: A massive benchmark and unified framework for LLM routing,

    H. Li, Y . Zhang, Z. Guo, C. Wang, S. Tang, Q. Zhang, Y . Chen, B. Qi, P. Ye, L. Bai, Z. Wang, and S. Hu, “LLMRouterBench: A massive benchmark and unified framework for LLM routing,” inFindings of ACL, 2026, arXiv:2601.07206

    arXiv 2026

  6. [6]

    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,”Transactions on Machine Learning Research (TMLR), 2024

    2024

  7. [7]

    RouteLLM: Learning to route LLMs from 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 from preference data,” inProc. ICLR, 2025

    2025

  8. [8]

    How much of the routing gap is real? Decomposing the router-to-oracle gap into reproducible specialist advantage and single- draw label noise,

    T.-R. Chen, “How much of the routing gap is real? Decomposing the router-to-oracle gap into reproducible specialist advantage and single- draw label noise,” 2026, arXiv:2607.03436 (companion paper)

    Pith/arXiv arXiv 2026

  9. [9]

    Enhancing LLM reasoning capabilities through brokered multi-expert reflection,

    T. Sheokand, G. Jain, A. Bahga, and V . K. Madisetti, “Enhancing LLM reasoning capabilities through brokered multi-expert reflection,”IEEE Access, vol. 13, pp. 67 993–68 019, 2025

    2025

  10. [10]

    Classifier cascades and trees for minimizing feature evaluation cost,

    Z. Xu, M. J. Kusner, K. Q. Weinberger, M. Chen, and O. Chapelle, “Classifier cascades and trees for minimizing feature evaluation cost,” Journal of Machine Learning Research, vol. 15, pp. 2113–2144, 2014

    2014

  11. [11]

    Classification with a reject option using a hinge loss,

    P. L. Bartlett and M. H. Wegkamp, “Classification with a reject option using a hinge loss,”Journal of Machine Learning Research, vol. 9, pp. 1823–1840, 2008

    2008

  12. [12]

    Machine learning with a reject option: A survey,

    K. Hendrickx, L. Perini, D. Van der Plas, W. Meert, and J. Davis, “Machine learning with a reject option: A survey,”Machine Learning, vol. 113, pp. 3073–3110, 2024

    2024

  13. [13]

    QoS-aware web service rec- ommendation by collaborative filtering,

    Z. Zheng, H. Ma, M. R. Lyu, and I. King, “QoS-aware web service rec- ommendation by collaborative filtering,”IEEE Transactions on Services Computing, vol. 4, no. 2, pp. 140–152, 2011

    2011

  14. [14]

    On the impact of deep neural network calibration on adaptive edge offloading for image classification,

    R. G. Pacheco, R. S. Couto, and O. Simeone, “On the impact of deep neural network calibration on adaptive edge offloading for image classification,”Journal of Network and Computer Applications, vol. 217, p. 103679, 2023

    2023

  15. [15]

    Deep neural networks meet computation offloading in mobile edge networks: Applications, taxonomy, and open issues,

    E. Mustafa, J. Shuja, F. Rehman, A. Riaz, M. Maray, M. Bilal, and M. K. Khan, “Deep neural networks meet computation offloading in mobile edge networks: Applications, taxonomy, and open issues,”Journal of Network and Computer Applications, vol. 226, p. 103886, 2024

    2024

  16. [16]

    Edge-AI: A systematic review on architectures, applications, and challenges,

    H. Gauttam, G. Nain, K. K. Pattanaik, and P. Mendes, “Edge-AI: A systematic review on architectures, applications, and challenges,” Journal of Network and Computer Applications, vol. 245, p. 104375, 2026

    2026

  17. [17]

    Self-consistency improves chain of thought reasoning in language models,

    X. Wang, J. Wei, D. Schuurmans, Q. Le, E. Chi, S. Narang, A. Chowdh- ery, and D. Zhou, “Self-consistency improves chain of thought reasoning in language models,” inProc. ICLR, 2023, arXiv:2203.11171

    Pith/arXiv arXiv 2023

  18. [18]

    Scaling LLM test-time compute optimally can be more effective than scaling parameters for reasoning,

    C. Snell, J. Lee, K. Xu, and A. Kumar, “Scaling LLM test-time compute optimally can be more effective than scaling parameters for reasoning,” inProc. ICLR, 2025, arXiv:2408.03314

    Pith/arXiv arXiv 2025

  19. [19]

    Ensemble-based uncertainty quantification for reliable large language model classification in social data applications,

    D. T. Farr, L. H. X. Ng, I. J. Cruickshank, N. Manzonelli, N. Clark, K. Starbird, N. D. Bastian, and J. West, “Ensemble-based uncertainty quantification for reliable large language model classification in social data applications,”IEEE Access, vol. 13, pp. 116 419–116 429, 2025. PREPRINT, JULY 2026 10

    2025

  20. [20]

    RouterEval: A comprehensive benchmark for routing LLMs to explore model-level scaling up in LLMs,

    Z. Huang, G. Ling, 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

    Pith/arXiv arXiv 2025

  21. [21]

    Beyond the leaderboard: A survey of the science of evaluation, benchmarking, and methodologies for large language models,

    S. Sheikhi, L. Lovén, and P. Kostakos, “Beyond the leaderboard: A survey of the science of evaluation, benchmarking, and methodologies for large language models,”IEEE Access, vol. 14, pp. 66 493–66 515, 2026

    2026

  22. [22]

    State of what art? A call for multi-prompt LLM eval- uation,

    M. Mizrahi, G. Kaplan, D. Malkin, R. Dror, D. Shahaf, and G. Stanovsky, “State of what art? A call for multi-prompt LLM eval- uation,”Transactions of the Association for Computational Linguistics, vol. 12, pp. 933–949, 2024

    2024

  23. [23]

    Make large language models efficient: A review,

    A. Mussa, Z. Tuimebayev, and M. Mansurova, “Make large language models efficient: A review,”IEEE Access, vol. 13, pp. 154 466–154 490, 2025

    2025

  24. [24]

    BrownoutServe: SLO-aware inference serving under bursty workloads for MoE-based LLMs,

    J. Hu, M. Xu, K. Ye, and C. Xu, “BrownoutServe: SLO-aware inference serving under bursty workloads for MoE-based LLMs,”IEEE Transac- tions on Computers, 2026, arXiv:2507.17133

    arXiv 2026

  25. [25]

    iGniter: Interference-aware GPU resource provisioning for predictable DNN inference in the cloud,

    F. Xu, J. Xu, J. Chen, L. Chen, R. Shang, Z. Zhou, and F. Liu, “iGniter: Interference-aware GPU resource provisioning for predictable DNN inference in the cloud,”IEEE Transactions on Parallel and Distributed Systems, vol. 34, no. 3, pp. 812–827, 2023

    2023

  26. [26]

    Best arm identification in multi-armed bandits,

    J.-Y . Audibert, S. Bubeck, and R. Munos, “Best arm identification in multi-armed bandits,” inProc. Conf. on Learning Theory (COLT), 2010, pp. 41–53

    2010

  27. [27]

    On the complexity of best- arm identification in multi-armed bandit models,

    E. Kaufmann, O. Cappé, and A. Garivier, “On the complexity of best- arm identification in multi-armed bandit models,”Journal of Machine Learning Research, vol. 17, no. 1, pp. 1–42, 2016

    2016

  28. [28]

    How can we know when language models know? On the calibration of language models for question answering,

    Z. Jiang, J. Araki, H. Ding, and G. Neubig, “How can we know when language models know? On the calibration of language models for question answering,”Transactions of the Association for Computational Linguistics, vol. 9, pp. 962–977, 2021

    2021