Pith. sign in

REVIEW 3 major objections 5 minor 64 references

Logits are All We Need to Adapt Closed Models

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper claims that aligning a black-box LLM with task-specific data reduces to label noise correction, and that reweighting logits alone is sufficient for adaptation.

desk verdict Genuinely useful empirical recipe, but the theory section doesn't match the algorithm and should not be taken at face value. read the letter →

arxiv 2502.06806 v4 pith:T4IYE46J submitted 2025-02-03 cs.LG cs.AIcs.CL

Gaurush Hiranandani , Haolun Wu , Subhojyoti Mukherjee , Sanmi Koyejo This is my paper · ORCID
classification cs.LGcs.AIcs.CL
keywords black-boxLLMadaptationtoken-levelprobabilityreweightinglogitaccesslabelnoisecorrectionautoregressivetransitionmatrixclosed-sourcemodelsdistributionshiftPluginmodel
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 aims to show that a closed large language model can be adapted to a task-specific distribution using token logits plus a small corpus of target examples, without touching weights or gradients. It reinterprets next-token prediction as classification under label noise: the closed model's broad-corpus probabilities act as noisy labels, and the target data act as clean labels. The proposed Plugin model is a small autoregressive reweighting network trained so that the final token distribution is the normalized element-wise product of the base model's probabilities and the reweighting vector; the paper claims this is sufficient to align generation with the task. The main theoretical result is a finite-time consistency bound showing that the estimated autoregressive transition matrix converges to its true value, and experiments across four datasets and three base models report consistent gains over prompting and logit-scaling baselines. If the claim is right, developers would not need fine-tuning APIs or exposed weights to customize closed models, only logit access.

What carries the argument

The machinery is the Plugin reweighting equation and its training objective. A small autoregressive model produces a per-token probability vector $r_t$; the combined distribution is the normalized element-wise product $p_t = b_t \odot r_t / \|b_t \odot r_t\|_1$, where $b_t$ is the frozen black-box model's softmax. The reweighting network is trained by sequence-level cross-entropy on target data with only its own parameters updated, and the theory treats $r_t$ as the diagonal of a class-conditional label-noise transition matrix $T_t$ that maps noisy (base-model) labels back to clean (target) labels. Theorem 1 gives the finite-time concentration argument that this estimated transition converges, which is the formal load-bearing step of the paper.

What would settle it

On a synthetic corpus where token A is systematically confused with token B in the base model, Plugin can only up-weight or down-weight A and B separately rather than swap their probabilities; if the target distribution requires that swap, Plugin's output cannot match it, and the diagonal assumption is directly falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that reweighting logits alone is sufficient for task adaptation. The paper casts the black-box model's per-token distribution as the noisy version of a target distribution and proposes to estimate an autoregressive transition vector from task-specific data. At inference, the Plugin model combines the frozen base probabilities $b_t$ with the reweighting model's output $r_t$ as $p_t = b_t \odot r_t / \|b_t \odot r_t\|_1$ and greedily decodes from $p_t$. Theorem 1 is stated as a high-probability bound on $L_t(\hat{\theta}_t) - L_t(\theta^*)$, showing that as the number of training tokens grows, the reweighting model's parameter estimate approaches the parameter governing the true autoregressive transition; this is the paper's basis for claiming the Plugin distribution aligns with the target. The empirical section supports the claim with improvements across BLEU, ROUGE, METEOR, and CIDEr on E2E NLG, Web NLG, CommonGen, and an Adidas product-description dataset, including settings where the base model has been biased by a distribution-shifted training set.

Load-bearing premise

The load-bearing premise is that a small network trained to reweight each token's probability independently is actually estimating the label-noise transition that separates the closed model's distribution from the target distribution; the paper asserts this equivalence when it restricts the transition to diagonal form but does not prove the converse.

Editorial extensions

If this is right

  • A developer with logit access and a small amount of task-specific text can adapt a closed model by training only a small reweighting network, with no weights, gradients, or vendor training data.
  • Plugin can wrap any prompt-based method: applying it on top of black-box discrete prompt learning improves that baseline on every dataset tested in the paper.
  • The reported gap between Plugin and a white-box fine-tuning method is small on several metrics, suggesting that logit reweighting recovers most of the observable task adaptation in low-data conditions.
  • Initializing the reweighting network with a small pretrained language model improves results, and a single transformer layer is reported to be near-optimal, so the overhead beyond the base model is modest in the paper's settings.
  • In distribution-shift evaluations, Plugin shifts generation toward the held-out target concept, such as raising Person-related Web NLG sentences from 17.99% to 71.34% and Woman-related CommonGen sentences from 10.37% to 31.92%.

Reading between the lines

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

  • A consequence the paper leaves implicit is that logit access could become a practical middle ground for closed-model deployment: vendors would expose an inference-only probability endpoint while keeping weights private, and the same reweighting recipe would transfer across tasks.
  • The diagonal restriction implies Plugin can only scale each token's probability, not move probability mass between tokens; tasks that require restructuring syntax or executing multi-step reasoning are therefore a likely ceiling, which matches the paper's own limitation statement.
  • A testable extension is to truncate the logit support to the top-k tokens and rerun Plugin: the degradation curve as k shrinks would quantify exactly how much access to full logits buys, and would tell vendors whether top-k APIs suffice.
  • One can also test the equivalence at the heart of the method by using a synthetic vocabulary with a known off-diagonal noise matrix; if the normalized elementwise product cannot represent that inverse transition, the consistency theorem's assumptions are not satisfied by the deployed algorithm.
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

3 major / 5 minor

Summary. The manuscript proposes Plugin, an inference-time adaptation method for closed-source LLMs that have logit access. A small autoregressive reweighting network r_t is trained on task-specific data, and generation uses the normalized elementwise product p_i = b_i ⊙ r_i / ||b_i ⊙ r_i||_1 of the base model's softmax b_i and the reweighting output r_i. The paper frames this as a label-noise correction problem: the base model is treated as a noisy label predictor, a transition matrix T_t is introduced, and the authors claim that estimating T_t and reweighting logits alone is sufficient for task adaptation. The theoretical section gives a finite-time bound for an M-estimator of a parameter θ* under a squared-loss surrogate, and the experiments compare Plugin with zero-shot prompting, in-context learning, a newly trained model, weighted combination, and TempNet on four datasets and three base models, with additional ablations, a wrapper experiment, human evaluation, and a LoRA comparison.

Significance. If the theoretical claim were correct, the paper would make a practically valuable contribution: logit access plus a small task set would allow closed-model adaptation without weights, gradients, or training data. The empirical package is a genuine strength: four datasets, three base models, seven metrics, five seeds, ablations on base-model quality and reweighting architecture, a distribution-shift analysis, a human evaluation, and a comparison with LoRA, with code availability stated. Unfortunately, the central theoretical bridge between the label-noise derivation and the implemented update in Eq. (4) is missing, so the paper currently establishes only that a product-of-softmax reweighting heuristic performs well in the studied settings; it does not establish the promised label-noise consistency guarantee. The claimed theoretical novelty is the main advertised contribution, and it is disconnected from the algorithm actually evaluated.

major comments (3)
  1. [Sec. 3.1, Eq. (4)] The derivation from the forward-corrected loss to the implemented reweighting is invalid. The matrix T_t in Eq. (3) is row-stochastic; restricting it to be diagonal forces T_t = I, so Eq. (3) reduces to the uncorrected loss. The text's statement that a diagonal transition matrix models "class-independent label noise" is also incorrect, because class-independent noise has nonzero off-diagonal probabilities. The normalized elementwise product p_i = b_i ⊙ r_i / ||b_i ⊙ r_i||_1 is nonlinear in b_i and is not of the form T_t^T b_i for any row-stochastic T_t; no argument is given that minimizing cross-entropy on Eq. (4) estimates T_t or applies its inverse. This missing step is load-bearing for the abstract's claim that reweighting logits alone is sufficient.
  2. [Sec. 5, Theorem 1 and Assumption 5.2] The theorem analyzes a different objective from the one used by Plugin. Assumption 5.2 asserts that the clipped cross-entropy loss is bounded by C|V|^2 (Y_t − f_I(θ*))^2, and the text then explicitly switches to "this squared loss" as ℓ_t(θ). Theorem 1 is therefore a finite-time M-estimation bound for squared-loss minimization; its statement never refers to b_t, r_t, p_t, or T_t, and it does not bound any divergence between Plugin's output distribution and the target distribution. An upper bound on cross-entropy by a squared loss does not imply that minimizing the squared loss minimizes the cross-entropy actually used in Eq. (5).
  3. [Sec. 5, Theorem 1] The consistency claim is largely assumed rather than derived. The theorem's assumptions already include the true parameter θ*, the transition function f_I, and a squared-loss bound; under those assumptions the proof is a standard M-estimation convergence argument. It does not establish that Plugin's autoregressive reweighting estimates the transition matrix of a label-noise process, because the mapping from the reweighting network r_t to the transition function f_I is never specified. The sentence after the theorem, claiming that accurate θ* estimation implies accurate estimation of T_t(θ*|F^{t−1}), therefore has no supporting statement in the theorem.
minor comments (5)
  1. [Table 3] The GPT2-XL zero-shot, ICL-1, and ICL-3 rows in Table 3 appear identical to the corresponding GPT2-XL entries in Table 2; this looks like a copy-paste error and should be corrected or explained.
  2. [Table 4] The GPT2-M ICL-1 row reports BLEU 0.0088 with standard deviation 0.054, which is not plausible for five seeds; several other ICL standard deviations in the same table appear to be an order of magnitude too large.
  3. [Abstract] The code link appears as the placeholder "this https URL" rather than an actual URL; the final version should include the link.
  4. [Sec. 7.4] The GPT-4o evaluation reports only aggregate percentages without specifying the number of generated samples, decoding parameters, or prompt repetitions; please add protocol details so the 17.99% versus 71.34% claim is reproducible.
  5. [Algorithm 1 / Eq. (5)] The index j is overloaded in Eq. (5): it is used both as a vocabulary coordinate and as the actual token position; please clarify that e_j denotes the one-hot vector of the token at position i.

Circularity Check

2 steps flagged · score 4.0 of 10

The theoretical bridge is definitional: §3's row-stochastic diagonal T_t is the identity, so Eq. (4)'s product reweighting cannot follow from label-noise correction, and Theorem 1 assumes the squared-loss/θ* link it is used to claim.

  1. self definitional [Section 3.1 and Section 4.1, Eq. (3)-(4)]
    "This matrix is row-stochastic but not necessarily symmetric. ... we impose the restriction that the transition matrix Tt is diagonal. ... The final token probability from the Plugin model is computed by normalizing the element-wise product of these probabilities: pi = bi ⊙ ri / ∥bi ⊙ ri∥1. (4) ... refining ri over iterations to approximate the transition vector governing label shifts in the target data."

    With the paper's own definition, a row-stochastic diagonal transition matrix is the identity, so the forward-corrected loss in Eq. (3) reduces to −log b and cannot produce any reweighting. The implemented p_i ∝ b_i r_i is not of the form T_t^T b for any non-identity row-stochastic T_t. The only way Eq. (4) can be called a transition-matrix correction is to define the learned reweighting vector r itself as the 'transition vector.' Thus the label-noise derivation does not generate Eq. (4); Eq. (4) is the reweighting ansatz, and the label-noise framing is attached to it by definition rather than derived from a noise process.

  2. other [Section 5, Assumption 5.2 and Theorem 1]
    "Assumption 5.2. We assume the cross-entropy loss (5) is clipped by ϵ > 0 and upper bounded as ℓclipped_t ≤ C|V|^2(Y_t − f_It(θ∗; xi, xj, F^{t−1}))^2 for any time t, where Y_t is the predicted token class, f_It determines the true class and satisfies Assumption 5.1, and C > 0 is a constant. ... Let the Plugin model minimize the loss ℓ_1(θ), ℓ_2(θ), · · ·, ℓ_t(θ). ... Since bθ_t = argmin_{θ∈Θ} Σ_{s=1}^t ℓ_s(θ), the Plugin model progressively refines its estimation of the unknown parameter θ∗."

    Under Assumption 5.2 the loss analyzed by Theorem 1 is replaced by a squared error to the true transition function f_It(θ∗; ...), so the theorem's conclusion that the empirical minimizer approaches L_t(θ∗) is a standard consistency statement with respect to the same θ∗ that was assumed to generate the loss. The theorem statement and proof never reference b_t, r_t, or the product distribution in Eq. (4), so the claimed guarantee that the Plugin output distribution matches the target distribution is not derived from the algorithm; it is effectively the content of Assumption 5.2 renamed as a theorem.

full rationale

The empirical evaluation of Plugin is self-contained and not circular: the reweighting model is trained on validation data and scored on held-out test data, and the comparisons (Tables 1-5, ablations, human evaluation) stand independently of the label-noise narrative. The circularity is confined to the paper's theoretical packaging. In §3 the noise transition matrix is defined as row-stochastic, then restricted to diagonal; any diagonal row-stochastic matrix is the identity, so Eq. (3) gives no correction. The algorithm of §4, p_i ∝ b_i⊙r_i, is not T_t^T b for any non-identity row-stochastic T_t; calling r_i the 'transition vector' defines the target of the derivation to be the algorithm itself. Theorem 1 then analyzes a squared loss whose true parameter θ* is written into Assumption 5.2, and never mentions b_t or r_t; its conclusion that θ* is recovered is a consistency statement about the assumed loss, not about the Plugin output distribution. These are two related instances of the central theoretical claim reducing to its own assumptions and definitions. Because the method has genuine independent empirical content, I score 4 rather than higher.

Assumptions & free parameters 2 free parameters · 5 assumptions · 1 invented entities

The paper's theoretical claim rests on several unverified modeling assumptions: that LLM-task differences are per-token label noise, that the transition matrix can be diagonal without changing the method, and that a squared-loss surrogate bounds the actual cross-entropy loss. The only genuinely fitted quantity for the empirical claim is the reweighting network itself, trained on validation data. The invented transition matrix T_t is a theoretical entity with no independent evidence.

free parameters (2)
  • Reweighting model parameters θ = Trained on validation data; final values not reported
    The reweighting network is the fitted object in Eq (4)-(5); its parameters are the mechanism that steers generation. This is the central fitted quantity.
  • Training hyperparameters (learning rate, weight decay, early stopping, architecture depth) = Cross-validated over learning rates {1e-5,...,5e-3}, weight decay {0.01,0.1,1,10}
    Section 7.1 says these are cross-validated on 40% of validation data; they determine the fitted reweighting model and are not predicted from first principles.
assumptions (5)
  • domain assumption Label-noise transition model: the LLM's broad training data is a class-conditional noisy version of the target task data, with per-token transition matrix T_t.
    Section 3 introduces this analogy; it is not empirically tested and is hard to reconcile with contextual, multi-token distribution shifts.
  • ad hoc to paper Diagonal restriction: T_t can be assumed diagonal so the transition matrix reduces to a vector modeled by a small autoregressive LM.
    Section 3.1 imposes this to reduce parameters; if T_t is row-stochastic and diagonal it is the identity, so the restriction contradicts the stated noise model and does not yield the product update.
  • ad hoc to paper Squared-loss bound: the clipped cross-entropy loss is upper bounded by C |V|^2 (Y_t - f_I(θ*))^2.
    Assumption 5.2 enables Theorem 1 but is stated without proof or empirical check; it is the only bridge from the algorithm's loss to the analyzed loss.
  • standard math Regularity and concentration: gradient/Hessian bounds, convexity, smoothness, and sub-Gaussian concentration hold for the transition function and loss.
    Assumptions 5.1 and B.1 are standard for M-estimation but are not verified for transformer-based reweighting models.
  • domain assumption Expressiveness: the black-box model is expressive enough to approximate the noisy label distribution p*(x̃_t|x_{t-1:1}).
    Section 3.1 borrows this from Patrini et al. (2017); it is needed for transition matrix estimation and is not tested here.
invented entities (1)
  • Autoregressive transition matrix T_t(θ*|F^{t-1}) with history-dependent transition function f_I
    purpose: Provides the theoretical rationale for reweighting logits as label-noise correction.
    The paper defines this construct and proves convergence for it, but the implemented Plugin never estimates or evaluates T_t directly; no falsifiable prediction from T_t is offered outside the paper's own assumptions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Logits are All We Need to Adapt Closed Models." pith.science (2026). https://pith.science/paper/T4IYE46J

@misc{pith2026250206806,
  author       = {Pith},
  title        = {Pith review of: Logits are All We Need to Adapt Closed Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T4IYE46J}},
  note         = {Machine review of arXiv:2502.06806}
}
read the original abstract

Many commercial Large Language Models (LLMs) are often closed-source, limiting developers to prompt tuning for aligning content generation with specific applications. While these models currently do not provide access to token logits, we argue that if such access were available, it would enable more powerful adaptation techniques beyond prompt engineering. In this paper, we propose a token-level probability reweighting framework that, given access to logits and a small amount of task-specific data, can effectively steer black-box LLMs toward application-specific content generation. Our approach views next-token prediction through the lens of supervised classification. We show that aligning black-box LLMs with task-specific data can be formulated as a label noise correction problem, leading to Plugin model -- an autoregressive probability reweighting model that operates solely on logits. We provide theoretical justification for why reweighting logits alone is sufficient for task adaptation. Extensive experiments with multiple datasets, LLMs, and reweighting models demonstrate the effectiveness of our method, advocating for broader access to token logits in closed-source models.

Figures

Figures reproduced from arXiv: 2502.06806 by the authors.

Figure 1
Figure 1. Inference phase of the Plugin model. The token proba￾bilities are a product of the probabilities from the black-box model and a reweighting model that denotes label transitioning. Despite these constraints, closed-source LLMs are widely adopted across applications ranging from travel itinerary generation to tax advice, with developers largely relying on prompt optimization to achieve domain-specific outputs. However… view at source ↗
Figure 2
Figure 2. Plugin with increasingly fine-tuned GPT2-M models on the E2E NLG dataset. Results demonstrate that as the quality of the base model improves, the performance of the Plugin improves. No plugin 1-layer2-layer4-layer8-layer12-layerGPT2 Small Choice of the reweighting module 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Value BLEU No plugin 1-layer2-layer4-layer8-layer12-layerGPT2 Small Choice of the reweighting module 0.00 0.10 0… view at source ↗
Figure 3
Figure 3. Performance of GPT2-M with varying reweighting model complexities on E2E NLG (BLEU, ROUGE-L). A single￾layer reweighting model yields significant gains, while additional layers degrade performance due to overfitting. Initializing with GPT2-Small as the reweighting model improves performance, demonstrating the benefits of leveraging small pretrained models. els, the architecture consists of a Transformer encoder with… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparison of the adaptation ability between the base model and Plugin on Adidas dataset. Plugin, enhanced with a reweighting model, generates text that better aligns with the “Adidas domain”. The bottom row illustrates token probabilities for key Adidas-related words …
Figure 5
Figure 5. Figure 5: Performance of applying a single-layer reweighting model across increasingly fine-tuned GPT2-M models on the three datasets. Results demonstrate consistent improvements introduced by our method regardless of the strength of the base model. No plugin 1-layer 2-layer 4-l…
Figure 6
Figure 6. Figure 6: Performance of GPT2-M with varying reweighting model complexities on the three datasets, measured by BLEU and Rouge-L. Results demonstrate that a single reweighting layer achieves significant improvements, while increasing the number of layers beyond this leads to perf…
Figure 7
Figure 7. Figure 7: Performance comparison of the weighting model architecture in Plugin. The transformer layer achieves the best performance with consistently higher means and narrower standard deviations. Shaded bands represent the standard deviation around the mean. C.7. Details for Ad…
Figure 8
Figure 8. Figure 8: Word Frequency of the Top-50 Words in Adidas Dataset. design is perfect for everyday wear, whether you’re running errands or hitting the gym. Plugin Model: Cushioned Angle Stripe Low-Cut Socks 6 Pairs are the perfect blend of style and comfort. The low-cut design gives…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 37 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Kaggle, 2023

    Adidas us retail products dataset. Kaggle, 2023. URL https://www.kaggle.com/datasets/whenamancodes/adidas-us-retail-products-dataset

    work page 2023

  3. [3]

    L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al

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

    arXiv 2023

  4. [4]

    Training a helpful and harmless assistant with reinforcement learning from human feedback

    Bai, Y., Jones, A., Ndousse, K., Askell, A., Chen, A., DasSarma, N., Drain, D., Fort, S., Ganguli, D., Henighan, T., et al. Training a helpful and harmless assistant with reinforcement learning from human feedback. arXiv preprint arXiv:2204.05862, 2022 a

    arXiv 2022

  5. [5]

    Constitutional ai: Harmlessness from ai feedback

    Bai, Y., Kadavath, S., Kundu, S., Askell, A., Kernion, J., Jones, A., Chen, A., Goldie, A., Mirhoseini, A., McKinnon, C., et al. Constitutional ai: Harmlessness from ai feedback. arXiv preprint arXiv:2212.08073, 2022 b

    arXiv 2022

  6. [6]

    and Lavie, A

    Banerjee, S. and Lavie, A. Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. In ACL workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pp.\ 65--72, 2005

    work page 2005

  7. [7]

    M., Netrapalli, P., and Sanghavi, S

    Chaudhuri, K., Kakade, S. M., Netrapalli, P., and Sanghavi, S. Convergence rates of active learning for maximum likelihood estimation. Advances in Neural Information Processing Systems, 28, 2015

    work page 2015

  8. [8]

    F., Leike, J., Brown, T., Martic, M., Legg, S., and Amodei, D

    Christiano, P. F., Leike, J., Brown, T., Martic, M., Legg, S., and Amodei, D. Deep reinforcement learning from human preferences. Advances in neural information processing systems, 30, 2017

    2017

Show all 64 references
  1. [9]

    Plug and play language models: A simple approach to controlled text generation

    Dathathri, S., Madotto, A., Lan, J., Hung, J., Frank, E., Molino, P., Yosinski, J., and Liu, R. Plug and play language models: A simple approach to controlled text generation. In International Conference on Learning Representations, 2020

    2020

  2. [10]

    Qlora: Efficient finetuning of quantized llms

    Dettmers, T., Pagnoni, A., Holtzman, A., and Zettlemoyer, L. Qlora: Efficient finetuning of quantized llms. Advances in Neural Information Processing Systems, 36, 2024

    2024

  3. [11]

    Black-box prompt learning for pre-trained language models

    Diao, S., Huang, Z., Xu, R., Li, X., Lin, Y., Zhou, X., and Zhang, T. Black-box prompt learning for pre-trained language models. arXiv preprint arXiv:2201.08531, 2022

    2022 arXiv

  4. [12]

    Automatic evaluation of machine translation quality using n-gram co-occurrence statistics

    Doddington, G. Automatic evaluation of machine translation quality using n-gram co-occurrence statistics. In International conference on Human Language Technology Research, pp.\ 138--145, 2002

    2002

  5. [13]

    A survey on in-context learning

    Dong, Q., Li, L., Dai, D., Zheng, C., Ma, J., Li, R., Xia, H., Xu, J., Wu, Z., Chang, B., et al. A survey on in-context learning. In Conference on Empirical Methods in Natural Language Processing, pp.\ 1107--1128, 2024

    2024

  6. [14]

    The llama 3 herd of models

    Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Letman, A., Mathur, A., Schelten, A., Yang, A., Fan, A., et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024

    2024 arXiv

  7. [15]

    Evaluating the state-of-the-art of end-to-end natural language generation: The e2e nlg challenge

    Du s ek, O., Novikova, J., and Rieser, V. Evaluating the state-of-the-art of end-to-end natural language generation: The e2e nlg challenge. Computer Speech & Language, 59: 0 123--156, 2020

    2020

  8. [16]

    M., and Sidford, A

    Frostig, R., Ge, R., Kakade, S. M., and Sidford, A. Competing with the empirical risk minimizer in a single pass. In Conference on learning theory, pp.\ 728--763. PMLR, 2015

    2015

  9. [17]

    Aligning llm agents by learning latent preference from user edits

    Gao, G., Taymanov, A., Salinas, E., Mineiro, P., and Misra, D. Aligning llm agents by learning latent preference from user edits. arXiv preprint arXiv:2404.15269, 2024

    2024 arXiv

  10. [18]

    Creating training corpora for nlg micro-planning

    Gardent, C., Shimorina, A., Narayan, S., and Perez-Beltrachini, L. Creating training corpora for nlg micro-planning. In Annual Meeting of the Association for Computational Linguistics, pp.\ 179--188, 2017

    2017

  11. [19]

    M., and Koyejo, S

    Hiranandani, G., Mathur, J., Narasimhan, H., Fard, M. M., and Koyejo, S. Optimizing black-box metrics with iterative example weighting. In International Conference on Machine Learning, pp.\ 4239--4249. PMLR, 2021

    2021

  12. [20]

    A tail inequality for quadratic forms of subgaussian random vectors

    Hsu, D., Kakade, S., Zhang, T., et al. A tail inequality for quadratic forms of subgaussian random vectors. Electronic Communications in Probability, 17, 2012

    2012

  13. [21]

    J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W

    Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021

    2021 arXiv

  14. [22]

    LLM -adapters: An adapter family for parameter-efficient fine-tuning of large language models

    Hu, Z., Wang, L., Lan, Y., Xu, W., Lim, E.-P., Bing, L., Xu, X., Poria, S., and Lee, R. LLM -adapters: An adapter family for parameter-efficient fine-tuning of large language models. In Conference on Empirical Methods in Natural Language Processing, pp.\ 5254--5276, 2023 a

    2023

  15. [23]

    Llm-adapters: An adapter family for parameter-efficient fine-tuning of large language models

    Hu, Z., Wang, L., Lan, Y., Xu, W., Lim, E.-P., Bing, L., Xu, X., Poria, S., and Lee, R. Llm-adapters: An adapter family for parameter-efficient fine-tuning of large language models. In Conference on Empirical Methods in Natural Language Processing, pp.\ 5254--5276, 2023 b

    2023

  16. [24]

    Calibrating long-form generations from large language models

    Huang, Y., Liu, Y., Thirukovalluru, R., Cohan, A., and Dhingra, B. Calibrating long-form generations from large language models. arXiv preprint arXiv:2402.06544, 2024

    2024 arXiv

  17. [25]

    P., Perelman, A., Ramesh, A., Clark, A., Ostrow, A., Welihinda, A., Hayes, A., Radford, A., et al

    Hurst, A., Lerer, A., Goucher, A. P., Perelman, A., Ramesh, A., Clark, A., Ostrow, A., Welihinda, A., Hayes, A., Radford, A., et al. Gpt-4o system card. arXiv preprint arXiv:2410.21276, 2024

    2024 arXiv

  18. [26]

    Calibration-tuning: Teaching large language models to know what they don’t know

    Kapoor, S., Gruver, N., Roberts, M., Pal, A., Dooley, S., Goldblum, M., and Wilson, A. Calibration-tuning: Teaching large language models to know what they don’t know. In Workshop on Uncertainty-Aware NLP, pp.\ 1--14, 2024

    2024

  19. [27]

    S., Reid, M., Matsuo, Y., and Iwasawa, Y

    Kojima, T., Gu, S. S., Reid, M., Matsuo, Y., and Iwasawa, Y. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35, 2022

    2022

  20. [28]

    O., Natarajan, N., Ravikumar, P

    Koyejo, O. O., Natarajan, N., Ravikumar, P. K., and Dhillon, I. S. Consistent binary classification with generalized performance metrics. Advances in neural information processing systems, 27, 2014

    2014

  21. [29]

    D., McCann, B., Keskar, N

    Krause, B., Gotmare, A. D., McCann, B., Keskar, N. S., Joty, S., Socher, R., and Rajani, N. F. Gedi: Generative discriminator guided sequence generation. In Findings of the Association for Computational Linguistics: EMNLP 2021, pp.\ 4929--4952, 2021

    2021

  22. [30]

    and Szepesv \'a ri, C

    Lattimore, T. and Szepesv \'a ri, C. Bandit algorithms. Cambridge University Press, 2020

    2020

  23. [31]

    The power of scale for parameter-efficient prompt tuning

    Lester, B., Al-Rfou, R., and Constant, N. The power of scale for parameter-efficient prompt tuning. In Conference on Empirical Methods in Natural Language Processing, pp.\ 3045--3059, 2021

    2021

  24. [32]

    S., and Hashimoto, T

    Li, X., Thickstun, J., Gulrajani, I., Liang, P. S., and Hashimoto, T. B. Diffusion-lm improves controllable text generation. Advances in Neural Information Processing Systems, 35: 0 4328--4343, 2022

    2022

  25. [33]

    Li, X. L. and Liang, P. Prefix-tuning: Optimizing continuous prompts for generation. In Annual Meeting of the Association for Computational Linguistics, pp.\ 4582--4597, 2021

    2021

  26. [34]

    Y., Zhou, W., Shen, M., Zhou, P., Bhagavatula, C., Choi, Y., and Ren, X

    Lin, B. Y., Zhou, W., Shen, M., Zhou, P., Bhagavatula, C., Choi, Y., and Ren, X. Commongen: A constrained text generation challenge for generative commonsense reasoning. In Findings of the Association for Computational Linguistics, pp.\ 1823--1840, 2020

    2020

  27. [35]

    Rouge: A package for automatic evaluation of summaries

    Lin, C.-Y. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pp.\ 74--81, 2004

    2004

  28. [36]

    and Och, F

    Lin, C.-Y. and Och, F. J. Automatic evaluation of machine translation quality using longest common subsequence and skip-bigram statistics. In Annual meeting of the association for computational linguistics, pp.\ 605--612, 2004

    2004

  29. [37]

    Detecting and correcting for label shift with black box predictors

    Lipton, Z., Wang, Y.-X., and Smola, A. Detecting and correcting for label shift with black box predictors. In International conference on machine learning, pp.\ 3122--3130. PMLR, 2018

    2018

  30. [38]

    A., and Choi, Y

    Liu, A., Sap, M., Lu, X., Swayamdipta, S., Bhagavatula, C., Smith, N. A., and Choi, Y. Dexperts: Decoding-time controlled text generation with experts and anti-experts. In Annual Meeting of the Association for Computational Linguistics, 2021

    2021

  31. [39]

    Liu, A., Han, X., Wang, Y., Tsvetkov, Y., Choi, Y., and Smith, N. A. Tuning language models by proxy. arXiv preprint arXiv:2401.08565, 2024

    2024 arXiv

  32. [40]

    Adapt in contexts: Retrieval-augmented domain adaptation via in-context learning

    Long, Q., Wang, W., and Pan, S. Adapt in contexts: Retrieval-augmented domain adaptation via in-context learning. In Conference on Empirical Methods in Natural Language Processing, pp.\ 6525--6542, 2023

    2023

  33. [41]

    Mix and match: Learning-free controllable text generationusing energy language models

    Mireshghallah, F., Goyal, K., and Berg-Kirkpatrick, T. Mix and match: Learning-free controllable text generationusing energy language models. In Annual Meeting of the Association for Computational Linguistics, pp.\ 401--415, 2022

    2022

  34. [42]

    S., and Nowak, R

    Mukherjee, S., Tripathy, A. S., and Nowak, R. Chernoff sampling for active testing and extension to active regression. In International Conference on Artificial Intelligence and Statistics, pp.\ 7384--7432. PMLR, 2022

    2022

  35. [43]

    and Javidi, T

    Naghshvar, M. and Javidi, T. Active sequential hypothesis testing. 2013

    2013

  36. [44]

    Consistent multiclass algorithms for complex performance measures

    Narasimhan, H., Ramaswamy, H., Saha, A., and Agarwal, S. Consistent multiclass algorithms for complex performance measures. In International Conference on Machine Learning, pp.\ 2398--2407. PMLR, 2015

    2015

  37. [45]

    Training language models to follow instructions with human feedback

    Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35: 0 27730--27744, 2022

    2022

  38. [46]

    Bleu: a method for automatic evaluation of machine translation

    Papineni, K., Roukos, S., Ward, T., and Zhu, W.-J. Bleu: a method for automatic evaluation of machine translation. In Annual meeting of the Association for Computational Linguistics, pp.\ 311--318, 2002

    2002

  39. [47]

    Making deep neural networks robust to label noise: A loss correction approach

    Patrini, G., Rozza, A., Krishna Menon, A., Nock, R., and Qu, L. Making deep neural networks robust to label noise: A loss correction approach. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 1944--1952, 2017

    1944

  40. [48]

    To cool or not to cool? temperature network meets large foundation models via dro

    Qiu, Z.-H., Guo, S., Xu, M., Zhao, T., Zhang, L., and Yang, T. To cool or not to cool? temperature network meets large foundation models via dro. arXiv preprint arXiv:2404.04575, 2024

    2024 arXiv

  41. [49]

    Language models are unsupervised multitask learners

    Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., Sutskever, I., et al. Language models are unsupervised multitask learners. OpenAI blog, 1 0 (8): 0 9, 2019

    2019

  42. [50]

    D., Ermon, S., and Finn, C

    Rafailov, R., Sharma, A., Mitchell, E., Manning, C. D., Ermon, S., and Finn, C. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36, 2024

    2024

  43. [51]

    Multitask prompted training enables zero-shot task generalization

    Sanh, V., Webson, A., Raffel, C., Bach, S., Sutawika, L., Alyafeai, Z., Chaffin, A., Stiegler, A., Raja, A., Dey, M., et al. Multitask prompted training enables zero-shot task generalization. In International Conference on Learning Representations, 2022

    2022

  44. [52]

    W., and Ghosh, S

    Shen, M., Das, S., Greenewald, K., Sattigeri, P., Wornell, G. W., and Ghosh, S. Thermometer: Towards universal calibration for large language models. In International Conference on Machine Learning, 2024

    2024

  45. [53]

    P., Hofmann, T., and Sch \"o lkopf, B

    Singh, S. P., Hofmann, T., and Sch \"o lkopf, B. The hessian perspective into the nature of convolutional neural networks. arXiv preprint arXiv:2305.09088, 2023

    2023 arXiv

  46. [54]

    H., Wu, J., Washington, C., Sadler, B

    Song, C. H., Wu, J., Washington, C., Sadler, B. M., Chao, W.-L., and Su, Y. Llm-planner: Few-shot grounded planning for embodied agents with large language models. In IEEE/CVF International Conference on Computer Vision, pp.\ 2998--3009, 2023

    2023

  47. [55]

    Training convolutional networks with noisy labels

    Sukhbaatar, S., Bruna, J., Paluri, M., Bourdev, L., and Fergus, R. Training convolutional networks with noisy labels. In International Conference on Learning Representations, 2015

    2015

  48. [56]

    On the duality between gradient transformations and adapters

    Torroba-Hennigen, L., Lang, H., Guo, H., and Kim, Y. On the duality between gradient transformations and adapters. arXiv preprint arXiv:2502.13811, 2025

    2025 arXiv

  49. [57]

    Ulmer, D., Gubri, M., Lee, H., Yun, S., and Oh, S. J. Calibrating large language models using their generations only. arXiv preprint arXiv:2403.05973, 2024

    2024 arXiv

  50. [58]

    Cider: Consensus-based image description evaluation

    Vedantam, R., Lawrence Zitnick, C., and Parikh, D. Cider: Consensus-based image description evaluation. In IEEE conference on computer vision and pattern recognition, pp.\ 4566--4575, 2015

    2015

  51. [59]

    W., Lester, B., Du, N., Dai, A

    Wei, J., Bosma, M., Zhao, V., Guu, K., Yu, A. W., Lester, B., Du, N., Dai, A. M., and Le, Q. V. Finetuned language models are zero-shot learners. In International Conference on Learning Representations, 2021

    2021

  52. [60]

    Transformers: State-of-the-art natural language processing

    Wolf, T. Transformers: State-of-the-art natural language processing. arXiv preprint arXiv:1910.03771, 2020

    1910 arXiv

  53. [61]

    and Klein, D

    Yang, K. and Klein, D. Fudge: Controlled text generation with future discriminators. In Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp.\ 3511--3535, 2021

    2021

  54. [62]

    A study on the calibration of in-context learning

    Zhang, H., Zhang, Y.-F., Yu, Y., Madeka, D., Foster, D., Xing, E., Lakkaraju, H., and Kakade, S. A study on the calibration of in-context learning. arXiv preprint arXiv:2312.04021, 2023

    2023 arXiv

  55. [63]

    Why transformers need adam: A hessian perspective

    Zhang, Y., Chen, C., Ding, T., Li, Z., Sun, R., and Luo, Z.-Q. Why transformers need adam: A hessian perspective. arXiv preprint arXiv:2402.16788, 2024

    2024 arXiv

  56. [64]

    On the calibration of large language models and alignment

    Zhu, C., Xu, B., Wang, Q., Zhang, Y., and Mao, Z. On the calibration of large language models and alignment. In Findings of the Association for Computational Linguistics: EMNLP 2023, pp.\ 9778--9795, 2023

    2023

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.