Pith. sign in

REVIEW 4 major objections 7 minor 57 references

ScaleOT: Privacy-utility-scalable Offsite-tuning with Dynamic LayerReplace and Selective Rank Compression

T0 review · 4 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read ScaleOT claims to deliver nearly lossless offsite tuning—plugging the returned adapter into the original model matches full fine-tuning—while the shared emulator is degraded enough to deter theft.

desk verdict Useful offsite-tuning method with strong plug-in numbers, but the privacy claim is an assertion, not a result—full emulator fine-tuning is never tested. read the letter →

arxiv 2412.09812 v1 pith:2TILCXYR submitted 2024-12-13 cs.CL cs.CR

Kai Yao , Zhaorui Tan , Tiandi Ye , Lichun Li , Yuan Zhao , Wenyan Liu , Wei Wang , Jianke Zhu This is my paper · ORCID
classification cs.CLcs.CR
keywords offsite-tuninglargelanguagemodelsprivacy-preservingfine-tuninglayerimportancereinforcementlearninglow-rankcompressionharmonizersemulator
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

This paper tries to establish a compression scheme for offsite-tuning that fixes two weaknesses of the standard approach: uniform layer dropping degrades the final adapted model, and knowledge distillation is too expensive. ScaleOT instead learns which transformer layers matter via reinforcement learning, replaces unimportant layers with lightweight harmonizers, and applies SVD low-rank compression only to attention matrices. The authors report that the resulting emulator can be fine-tuned by a data owner and the returned adapter, plugged back into the full model, performs about as well as full fine-tuning—sometimes slightly better—while the emulator itself scores much worse, which they take as model privacy. The practical payoff would be a cheap, adjustable way to let third parties fine-tune large models without exposing the weights or the data.

What carries the argument

The load-bearing mechanism is Dynamic LayerReplace combined with Selective Rank Compression. Dynamic LayerReplace learns a per-layer importance score $s_i$ via reinforcement learning: candidate networks are formed by replacing half the layers in each of $N_g = 4$ groups with harmonizers (lightweight low-rank FFNs), rewards are $r_j = e^{-L_j} - \frac{1}{N_c}\sum_t e^{-L_t}$, and scores update by $s_i \leftarrow s_i + r_j \sigma(s_i)(1-\sigma(s_i))$ for layers in the sampled set. The emulator is then composed of harmonizers for the least important layers and, for the remaining frozen layers, SVD-compressed attention matrices with rank reduction ratio $\beta$; the adapter is the top $N_a/N_g$ layers per group. This machinery creates a deliberate asymmetry: attention rank compression degrades the emulator's own fine-tuned accuracy quickly while the plug-in performance declines slowly, which is what turns compression into a privacy lever rather than just a utility cost.

What would settle it

Give an adversary the emulator, its harmonizers, and the returned adapter, allow distillation on a public proxy dataset, and check whether the recovered model's accuracy approaches full fine-tuning; if it does while the emulator's own zero-shot and fine-tuned scores remain low, the privacy claim collapses.

Watch

Extended reading notes

Core claim

The central claim is that a lossy emulator built by replacing low-importance transformer layers with lightweight harmonizers chosen by a reinforcement-learning importance score, and then applying SVD low-rank compression only to multi-head attention matrices, can be fine-tuned by a data owner whose returned adapter restores the original model to nearly full-fine-tuning accuracy. On multiple benchmarks the plug-in performance matches or slightly exceeds direct fine-tuning, while the emulator's own zero-shot and fine-tuned scores fall substantially. The authors read this performance gap as model privacy: a downstream user who keeps the emulator gets a weaker model, and is therefore encouraged to return the adapter and use the plugged-in original.

Load-bearing premise

That a noticeably weaker emulator, measured by benchmark accuracy, cannot be used by an adversary to extract or reconstruct a usable model from the returned adapter.

Editorial extensions

If this is right

  • Plug-in accuracy roughly matches full fine-tuning on the tested QA and language-modeling tasks, and on several benchmarks it slightly exceeds it, so downstream users lose little by using the returned adapter.
  • Because emulator zero-shot and fine-tuned performance drop while plug-in stays high, the shared artifact satisfies the three conditions the paper defines for effective offsite tuning: zero-shot below plug-in, emulator fine-tuning below plug-in, and plug-in close to full fine-tuning.
  • The two knobs $\alpha$ (harmonizer replacement ratio) and $\beta$ (rank reduction ratio) give a privacy-utility frontier, letting a model owner issue emulators of different scales rather than one fixed compressed model.
  • ScaleOT composes with parameter-efficient methods: adding LoRA (rank 4) or adapters (dimension 64) cuts trainable parameters to below 2 million while preserving plug-in perplexity.
  • Compression requires training only about 1–2% of the model's parameters, which the paper argues makes offsite tuning practical for models in the 7-billion-parameter range without knowledge distillation.

Reading between the lines

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

  • The paper's privacy claim rests on the emulator's weak task scores; it does not run extraction, distillation, or inversion attacks, so the actual privacy guarantee under an adaptive adversary is undetermined.
  • The finding that tuning only important layers can beat full fine-tuning suggests the same reinforcement-learned importance scores could be repurposed for memory-efficient fine-tuning of larger models, a connection the authors note but do not develop into a method.
  • Selective Rank Compression's choice to compress attention but not feedforward weights is an empirical asymmetry; testing whether the same asymmetry transfers to other architectures, such as vision or multimodal transformers, would be a natural check.
  • The returned adapter itself may carry information about the original layers; analyzing whether the adapter can be combined with the emulator to reconstruct a higher-performing model would tighten the privacy argument.
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 / 7 minor

Summary. The paper proposes ScaleOT, an offsite-tuning framework for large language models that combines three components: a reinforcement-learning-based layer-importance estimator (Dynamic LayerReplace), lightweight 'harmonizer' networks that replace low-importance layers, and Selective Rank Compression (SRC), which applies SVD-based low-rank approximation to the MHSA weights of the retained layers. The framework is intended to generate emulators of varying sizes and compression levels so that data owners can fine-tune an adapter on the emulator and return it to the model owner, achieving plug-in performance close to full fine-tuning while keeping the emulator too weak to be a usable substitute. Experiments are reported on GPT-2-XL, OPT-1.3B/6.7B, and LLaMA-7B across WikiText and eight question-answering benchmarks, with ablations over the replacement ratio α, the rank compression ratio β, and the number of adapted layers Na.

Significance. If the utility results hold, ScaleOT is a useful step toward practical offsite tuning: it avoids the expensive knowledge-distillation phase of vanilla OT, provides a tunable privacy-utility knob through α and β, and is shown to be orthogonal to adapter-based PEFT methods. The empirical coverage across model families and tasks is a genuine strength, and the observation that MHSA rank compression degrades emulator performance faster than plug-in performance is interesting and potentially actionable. However, the paper's central privacy claim is not yet supported by evidence against an actual adversary. The authors operationalize privacy solely as a performance gap under the cooperative adapter-only tuning protocol and do not evaluate the natural attack of fine-tuning the full emulator, so the 'better model privacy' half of the abstract is currently an inference, not an established result. No code or model release is mentioned, which further limits reproducibility.

major comments (4)
  1. [§3, §4.2, Table 1] The privacy claim rests entirely on the observation that the fine-tuned emulator underperforms the plug-in model (e.g., Table 1: GPT2-XL Emulator FT 45.0 vs Plug-in 49.3; OPT-1.3B 46.2 vs 49.9). This observation is obtained under the cooperative protocol that fine-tunes only the adapter layers. An adversary who obtains the emulator is not obliged to follow this protocol: she can fine-tune the entire emulator, including the harmonizers and the SVD-compressed MHSA layers, optionally initializing from the returned adapter. The SVD factors are merely initializations, and full-rank fine-tuning can undo the rank restriction. The statement in Section 4.2 that SRC 'certifies privacy protection' is therefore unsupported. Please define a threat model and evaluate at least the full-emulator fine-tuning attack (and ideally a distillation or model-extraction baseline) for GPT2-XL and OPT-1.3B; if the gap closes under this attack, the 'better model privacy' claim should be revised or qualified.
  2. [§5, Tables 1–3, Figs. 3–4] No variance, confidence intervals, or repeated runs are reported anywhere in the empirical evaluation. The differences that support 'nearly lossless' and the α/β trends are often only 1–3 points on task averages (e.g., Table 1: ScaleOT Plug-in 49.3 vs Full FT 49.9 for GPT2-XL), which is within the typical run-to-run noise of fine-tuning experiments. Please report mean ± std over at least three random seeds for the main tables and for the Fig. 4 trade-off curves, or otherwise justify that seed variance is negligible for these settings.
  3. [§4.1, Eq. (5)] The importance-score update in Eq. (5) is a heuristic reward-weighted gradient step rather than a standard policy-gradient or otherwise principled RL update, and no objective, convergence guarantee, or analysis is provided. Because the entire Dynamic LayerReplace procedure and the subsequent emulator creation depend on these scores, the paper should provide evidence that the scores are reliable and stable: for example, report the RL reward/loss curve over training, an ablation over the number of candidate networks Nc, and a stability check such as the correlation between importance scores from independent training runs or agreement with an oracle importance measure like layer-wise gradient norms.
  4. [§5, Table 2] The 'nearly lossless offsite tuning compared with full fine-tuning' claim cannot be verified for the large models: Table 2 reports zero-shot and plug-in results for OPT-6.7B and LLaMA-7B but no full fine-tuning numbers. In addition, the CRaSh baseline is evaluated only on the large models, so the medium-model comparison in Table 1 omits a strong recent baseline. Please add the missing full-FT baselines for the large models and CRaSh results on GPT2-XL and OPT-1.3B, or explicitly state that those runs were not performed and qualify the lossless claim accordingly.
minor comments (7)
  1. [Preliminary] The word 'definations' should be 'definitions'.
  2. [Fig. 3 caption] The caption 'with varies β' should read 'with various β'.
  3. [Eq. (1)] The symbol π_i is used both for a distribution and for the sampled probability p_i; please use distinct notation, for example π_i for the policy and p_i for the realized sample.
  4. [Eqs. (8)–(10)] The relationship between k = Na/Ng and κ = n×α/Ng is not explained, and it is not immediately obvious that the harmonizer index set φ_H is always a subset of φ_E for arbitrary α and Na; please clarify the intended inequalities and ranges.
  5. [References] The reference 'et al., R. P. 2024' is malformed; it should be 'Pan, R., et al. 2024' (the LISA paper).
  6. [Table 3] In Table 3 the column headers 'Emu. FT ↑' and 'Plug-in ↓' are confusing because lower perplexity is better; the arrows seem to denote desired directions for different columns, so please clarify.
  7. [§4.2] The phrase 'certify privacy protection' is too strong for a performance-gap observation; consider replacing it with a more operational statement such as 'increases the gap under the protocol described.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the utility and privacy numbers are measured outcomes, not fitted targets, and the only self-citations are motivational rather than load-bearing.

full rationale

The paper's derivation chain is self-contained and benchmarked against external baselines (OT, CRaSh, and full fine-tuning). The RL importance scores, harmonizer training, and SVD rank compression are fixed procedures evaluated on held-out benchmarks; the resulting emulator, plug-in, and full-fine-tuning numbers are measurements, not predictions recovered from fitted parameters. The default settings α=0.25 and β=0.8 are selected from the paper's own sweep, but this is ordinary hyperparameter choice rather than a statistically forced 'prediction,' and the paper does not claim to derive the sweep outcomes from the defaults. The privacy claim is indeed weak—'privacy' is operationalized as the emulator-FT-vs-plug-in gap (Section 3, condition 2) and no adversarial full fine-tuning of the emulator is evaluated—but that is an unsupported empirical assumption, not a circular reduction: SRC is defined as rank compression of MHSA layers, and whether it widens the gap is an empirical finding (Fig. 3). The only self-citations (Yao et al. 2024) support the general observation that layer importance varies and are not load-bearing for the paper's central claims. No circular step can be exhibited.

Assumptions & free parameters 6 free parameters · 4 assumptions · 1 invented entities

The framework rests on several empirical choices (alpha, beta, Ng, Nc, Na, harmonizer rank) and on the unvalidated assumption that a degraded emulator equals privacy. SVD low-rank approximation is standard. The RL update rule is a heuristic without a stated objective.

free parameters (6)
  • alpha (layer replacement ratio) = 0.25 (default)
    Proportion of layers replaced with harmonizers; chosen empirically to balance privacy and utility, and varied in Fig. 4.
  • beta (rank compression ratio) = 0.8 (default)
    Rank reduction ratio applied to MHSA weights via SVD; chosen empirically as a favorable privacy-utility operating point in Fig. 3 and Fig. 4.
  • Ng (number of adjacent-layer groups) = 4
    Set empirically to stabilize RL training by preventing selection of too many adjacent layers.
  • Nc (number of sampled candidate networks) = 3
    Number of candidate networks sampled per RL update; set empirically.
  • Na (number of adapted layers) = 4 (about 10% of parameters)
    Set to match the OT baseline and observed to saturate at Na=4 in Fig. 5.
  • Harmonizer hidden rank = 64 (medium), 256 (large)
    Rank of the low-rank FFN harmonizers; chosen by model size without systematic sweep.
assumptions (4)
  • standard math Eckart-Young-Mirsky theorem guarantees SVD gives the optimal rank-r approximation in Frobenius/spectral norm.
    Used in Section 4.2 to justify the rank-r approximation of MHSA weight matrices.
  • domain assumption Privacy is equivalent to a performance gap between the emulator and the plug-in full model.
    Stated in Section 3 as the offsite-tuning requirements and used throughout to claim 'better model privacy' and 'certify privacy protection', but no adversary model or attack evaluation is provided.
  • domain assumption Layer importance scores estimated on a pretraining corpus (Pile) transfer to downstream QA tasks.
    The importance scores and harmonizers are trained on Pile in Section 5.2, then used to build emulators for WikiText and eight QA benchmarks, assuming the learned layer importance generalizes.
  • ad hoc to paper The RL update rule in Eq. (5) converges to meaningful layer importance scores.
    The update is heuristic: rewards are computed from sampled candidate networks and applied with a sigmoid-scaled increment; no objective function, convergence guarantee, or theoretical justification is given in Section 4.1.
invented entities (1)
  • Harmonizer
    purpose: A lightweight low-rank FFN inserted in place of less-important transformer layers to preserve semantic consistency between remaining layers.
    A new trainable module proposed for compression; its effectiveness is demonstrated only by the paper's own experiments, with no external validation or theoretical analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ScaleOT: Privacy-utility-scalable Offsite-tuning with Dynamic LayerReplace and Selective Rank Compression." pith.science (2026). https://pith.science/paper/2TILCXYR

@misc{pith2026241209812,
  author       = {Pith},
  title        = {Pith review of: ScaleOT: Privacy-utility-scalable Offsite-tuning with Dynamic LayerReplace and Selective Rank Compression},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2TILCXYR}},
  note         = {Machine review of arXiv:2412.09812}
}
read the original abstract

Offsite-tuning is a privacy-preserving method for tuning large language models (LLMs) by sharing a lossy compressed emulator from the LLM owners with data owners for downstream task tuning. This approach protects the privacy of both the model and data owners. However, current offsite tuning methods often suffer from adaptation degradation, high computational costs, and limited protection strength due to uniformly dropping LLM layers or relying on expensive knowledge distillation. To address these issues, we propose ScaleOT, a novel privacy-utility-scalable offsite-tuning framework that effectively balances privacy and utility. ScaleOT introduces a novel layerwise lossy compression algorithm that uses reinforcement learning to obtain the importance of each layer. It employs lightweight networks, termed harmonizers, to replace the raw LLM layers. By combining important original LLM layers and harmonizers in different ratios, ScaleOT generates emulators tailored for optimal performance with various model scales for enhanced privacy protection. Additionally, we present a rank reduction method to further compress the original LLM layers, significantly enhancing privacy with negligible impact on utility. Comprehensive experiments show that ScaleOT can achieve nearly lossless offsite tuning performance compared with full fine-tuning while obtaining better model privacy.

Figures

Figures reproduced from arXiv: 2412.09812 by the authors.

Figure 1
Figure 1. Comparison of layerwise compression strategies. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Comparison in various tuning methods. (a) Fine-tuning requires access to full model parameters and necessitates the [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. Joint effect of α and β in emulator generation. while keeping the less ones within the LLM unchanged, thus forming A = {mi}i∈ϕA and E = {mi}i∈ϕE . Subsequently, the indices set of harmonizers is denoted as follows: ϕH = {i where si < sgj ,κ, i ∈ gj}, (10) where κ = n×α Ng , s gj ,κ is the κ-th largest importance score in j-th group, ϕH ∈ ϕE. The emulator E ∗ can be formed as: E ∗ = {hi}i∈ϕH ∪ {SRC(mi , β)}i∈ϕE ∧i /∈… view at source ↗
Figures from the paper (2 more)
Figure 3
Figure 3. Figure 3: Rank Compression Study with varies β on multi￾head self-attention (MHSA) layer and a Feedforward Net￾work (FFN) layer in Transformer Block on WikiText dataset. Rank Compression on Specific Modules. A typical Trans￾former (Vaswani et al. 2017) layer is composed of a mul…
Figure 5
Figure 5. Figure 5: Number of adapted layers in ScaleOT. creased, particularly in configurations that included the feed￾forward network (FFN), where a linear relationship was ev￾ident. In contrast, for MHSA configurations, within a range of 0.6 to 1, emulator FT performance showed an expo…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 34 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    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 word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Berant, J.; Chou, A.; Frostig, R.; and Liang, P. 2013. Semantic parsing on freebase from question-answer pairs. In Conference on Empirical Methods in Natural Language Processing, 1533--1544

    work page 2013

  4. [4]

    Bisk, Y.; Zellers, R.; Gao, J.; Choi, Y.; et al. 2020. Piqa: Reasoning about physical commonsense in natural language. In AAAI, volume 34, 7432--7439

    work page 2020

  5. [5]

    D.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; et al

    Brown, T.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J. D.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; et al. 2020. Language models are few-shot learners. Advances in Neural Information Processing Systems, 33: 1877--1901

    2020

  6. [6]

    J.; Yu, W.; Zhao, J.; and Lam, K.-Y

    Chua, T. J.; Yu, W.; Zhao, J.; and Lam, K.-Y. 2023. FedPEAT: Convergence of Federated Learning, Parameter-Efficient Fine Tuning, and Emulator Assisted Tuning for Artificial Intelligence Foundation Models with Mobile Edge Computing. arXiv preprint arXiv:2310.17491

    arXiv 2023

  7. [7]

    Clark, P.; Cowhey, I.; Etzioni, O.; Khot, T.; Sabharwal, A.; Schoenick, C.; and Tafjord, O. 2018. Think you have solved question answering? try arc, the ai2 reasoning challenge. arXiv preprint arXiv:1803.05457

    arXiv 2018

  8. [8]

    Dai, W.; Li, J.; Li, D.; Tiong, A. M. H.; Zhao, J.; Wang, W.; Li, B.; Fung, P. N.; and Hoi, S. 2024. Instructblip: Towards general-purpose vision-language models with instruction tuning. Advances in Neural Information Processing Systems, 36

    2024

Show all 57 references
  1. [9]

    Devlin, J.; Chang, M.-W.; Lee, K.; and Toutanova, K. 2019. BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language T...

    2019

  2. [10]

    Dong, Q.; Li, L.; Dai, D.; Zheng, C.; Ma, J.; Li, R.; Xia, H.; Xu, J.; Wu, Z.; Chang, B.; Sun, X.; and Sui, Z. 2024. A Survey on In-context Learning. In Conference on Empirical Methods in Natural Language Processing, 1107--1128

    2024

  3. [11]

    Du, Z.; Qian, Y.; Liu, X.; Ding, M.; Qiu, J.; Yang, Z.; and Tang, J. 2022. GLM: General Language Model Pretraining with Autoregressive Blank Infilling. In Association for Computational Linguistics, 320--335

    2022

  4. [12]

    Eckart, C.; and Young, G. 1936. The approximation of one matrix by another of lower rank. Psychometrika, 1(3): 211--218

    1936

  5. [14]

    Fan, T.; Kang, Y.; Ma, G.; Chen, W.; Wei, W.; Fan, L.; and Yang, Q. 2023. Fate-llm: A industrial grade federated learning framework for large language models. arXiv preprint arXiv:2310.10049

    2023 arXiv

  6. [15]

    Frankle, J.; and Carbin, M. 2018. The lottery ticket hypothesis: Finding sparse, trainable neural networks. arXiv preprint arXiv:1803.03635

    2018 arXiv

  7. [16]

    Gao, L.; Biderman, S.; Black, S.; Golding, L.; Hoppe, T.; Foster, C.; Phang, J.; He, H.; Thite, A.; Nabeshima, N.; et al. 2020. The pile: An 800gb dataset of diverse text for language modeling. arXiv preprint arXiv:2101.00027

    2020 arXiv

  8. [17]

    Han, S.; Mao, H.; and Dally, W. J. 2016. Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding. In International Conference on Learning Representations

    2016

  9. [18]

    Hinton, G.; Vinyals, O.; and Dean, J. 2015. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531

    2015 arXiv

  10. [19]

    Hoefler, T.; Alistarh, D.; Ben-Nun, T.; Dryden, N.; and Peste, A. 2021. Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks. Journal of Machine Learning Research, 22(241): 1--124

    2021

  11. [20]

    T.; Zhang, C.; LI, Z.; Li, B.; and Wang, Z

    Hong, J.; Wang, J. T.; Zhang, C.; LI, Z.; Li, B.; and Wang, Z. 2024. DP-OPT: Make Large Language Model Your Privacy-Preserving Prompt Engineer. In International Conference on Learning Representations

    2024

  12. [21]

    Houlsby, N.; Giurgiu, A.; Jastrzebski, S.; Morrone, B.; De Laroussilhe, Q.; Gesmundo, A.; Attariyan, M.; and Gelly, S. 2019. Parameter-efficient transfer learning for NLP. In International Conference on Machine Learning, 2790--2799

    2019

  13. [22]

    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. 2022. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations

    2022

  14. [23]

    Jacob, B.; Kligys, S.; Chen, B.; Zhu, M.; Tang, M.; Howard, A.; Adam, H.; and Kalenichenko, D. 2018. Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    2018

  15. [24]

    C.; Lo, W.-Y.; et al

    Kirillov, A.; Mintun, E.; Ravi, N.; Mao, H.; Rolland, C.; Gustafson, L.; Xiao, T.; Whitehead, S.; Berg, A. C.; Lo, W.-Y.; et al. 2023. Segment anything. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 4015--4026

    2023

  16. [25]

    Kuang, W.; Qian, B.; Li, Z.; Chen, D.; Gao, D.; Pan, X.; Xie, Y.; Li, Y.; Ding, B.; and Zhou, J. 2024. FederatedScope-LLM: A Comprehensive Package for Fine-tuning Large Language Models in Federated Learning. In Conference on Knowledge Discovery and Data Mining, 5260--5271

    2024

  17. [26]

    Lai, G.; Xie, Q.; Liu, H.; Yang, Y.; and Hovy, E. 2017. Race: Large-scale reading comprehension dataset from examinations. In Conference on Empirical Methods in Natural Language Processing, 785--794

    2017

  18. [27]

    Lan, Z.; Chen, M.; Goodman, S.; Gimpel, K.; Sharma, P.; and Soricut, R. 2020. Albert: A lite bert for self-supervised learning of language representations. In International Conference on Learning Representations

    2020

  19. [28]

    Li, L.; Ma, R.; Guo, Q.; Xue, X.; and Qiu, X. 2020. BERT-ATTACK: Adversarial Attack Against BERT Using BERT . In Conference on Empirical Methods in Natural Language Processing, 6193--6202

    2020

  20. [29]

    Liu, H.; Li, C.; Wu, Q.; and Lee, Y. J. 2024 a . Visual instruction tuning. Advances in Neural Information Processing Systems, 36

    2024

  21. [30]

    Liu, H.; Liu, Z.; Tang, R.; Yuan, J.; Zhong, S.; Chuang, Y.-N.; Li, L.; Chen, R.; and Hu, X. 2024 b . LoRA-as-an-Attack! Piercing LLM Safety Under The Share-and-Play Scenario. arXiv preprint arXiv:2403.00108

    2024 arXiv

  22. [31]

    Lv, X.; Zhang, P.; Li, S.; Gan, G.; and Sun, Y. 2023. Lightformer: Light-weight transformer using svd-based weight transfer and parameter sharing. In Findings of the Association for Computational Linguistics, 10323--10335

    2023

  23. [32]

    Merity, S.; Xiong, C.; Bradbury, J.; and Socher, R. 2017. Pointer Sentinel Mixture Models. In International Conference on Learning Representations

    2017

  24. [33]

    Michel, P.; Levy, O.; and Neubig, G. 2019. Are sixteen heads really better than one? Advances in Neural Information Processing Systems, 32: 14014--14024

    2019

  25. [34]

    Mihaylov, T.; Clark, P.; Khot, T.; and Sabharwal, A. 2018. Can a suit of armor conduct electricity? a new dataset for open book question answering. In Conference on Empirical Methods in Natural Language Processing

    2018

  26. [35]

    Murty, S.; Sharma, P.; Andreas, J.; and Manning, C. D. 2023. Characterizing intrinsic compositionality in transformers with Tree Projections. In International Conference on Learning Representations

    2023

  27. [36]

    C.; Ding, M.; Pathirana, P

    Nguyen, D. C.; Ding, M.; Pathirana, P. N.; Seneviratne, A.; Li, J.; and Poor, H. V. 2021. Federated learning for internet of things: A comprehensive survey. IEEE Communications Surveys & Tutorials, 23(3): 1622--1658

    2021

  28. [37]

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

    2022

  29. [38]

    Pan, R.; Liu, X.; Diao, S.; Pi, R.; Zhang, J.; Han, C.; and Zhang, T. 2024. LISA: Layerwise Importance Sampling for Memory-Efficient Large Language Model Fine-Tuning. arXiv preprint arXiv:2403.17919

    2024 arXiv

  30. [39]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning transferable visual models from natural language supervision. In International Conference on Machine Learning, 8748--8763

    2021

  31. [40]

    W.; Xu, T.; Brockman, G.; McLeavey, C.; and Sutskever, I

    Radford, A.; Kim, J. W.; Xu, T.; Brockman, G.; McLeavey, C.; and Sutskever, I. 2023. Robust speech recognition via large-scale weak supervision. In International Conference on Machine Learning, 28492--28518. PMLR

    2023

  32. [41]

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

    2019

  33. [42]

    Sajjad, H.; Dalvi, F.; Durrani, N.; and Nakov, P. 2023. On the effect of dropping layers of pre-trained transformer models. Computer Speech & Language, 77: 101429

    2023

  34. [43]

    Sanh, V.; Debut, L.; Chaumond, J.; and Wolf, T. 2019. DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108

    2019 arXiv

  35. [44]

    Tan, Z.; Yang, X.; Wang, Q.; Nguyen, A.; and Huang, K. 2024. Interpret Your Decision: Logical Reasoning Regularization for Generalization in Visual Classification. In Advances in Neural Information Processing Systems

    2024

  36. [45]

    Touvron, H.; Martin, L.; Stone, K.; Albert, P.; Almahairi, A.; Babaei, Y.; Bashlykov, N.; Batra, S.; Bhargava, P.; Bhosale, S.; et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288

    2023 arXiv

  37. [46]

    N.; Kaiser, L

    Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A. N.; Kaiser, L. u.; and Polosukhin, I. 2017. Attention is All you Need. In Advances in Neural Information Processing Systems, volume 30

    2017

  38. [47]

    Y.; Guu, K.; Yu, A

    Wei, J.; Bosma, M.; Zhao, V. Y.; Guu, K.; Yu, A. W.; Lester, B.; Du, N.; Dai, A. M.; and Le, Q. V. 2022. Finetuned Language Models are Zero-Shot Learners. In International Conference on Learning Representations

    2022

  39. [48]

    F.; and Gardner, M

    Welbl, J.; Liu, N. F.; and Gardner, M. 2017. Crowdsourcing Multiple Choice Science Questions. In EMNLP Workshop, 94--106. Association for Computational Linguistics

    2017

  40. [49]

    Wies, N.; Levine, Y.; and Shashua, A. 2024. The learnability of in-context learning. Advances in Neural Information Processing Systems, 36

    2024

  41. [50]

    W.; Li, M.; Kornblith, S.; Roelofs, R.; Lopes, R

    Wortsman, M.; Ilharco, G.; Kim, J. W.; Li, M.; Kornblith, S.; Roelofs, R.; Lopes, R. G.; Hajishirzi, H.; Farhadi, A.; and Namkoong, H. 2022. Robust fine-tuning of zero-shot models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 7959--7971

    2022

  42. [51]

    Xiao, G.; Lin, J.; and Han, S. 2023. Offsite-tuning: Transfer learning without full model. arXiv preprint arXiv:2302.04870

    2023 arXiv

  43. [52]

    Yao, K.; Gao, P.; Li, L.; Zhao, Y.; Wang, X.; Wang, W.; and Zhu, J. 2024. Layer-wise Importance Matters: Less Memory for Better Performance in Parameter-efficient Fine-tuning of Large Language Models. In Findings of the Association for Computational Linguistics: EMNLP 2024, 1977--1992

    2024

  44. [53]

    Ye, T.; Chen, C.; Wang, Y.; Li, X.; and Gao, M. 2024. BapFL: You can Backdoor Personalized Federated Learning. ACM Trans. Knowl. Discov. Data , 18(7): 166

    2024

  45. [54]

    Zellers, R.; Holtzman, A.; Bisk, Y.; Farhadi, A.; and Choi, Y. 2019. HellaSwag: Can a Machine Really Finish Your Sentence? In Association for Computational Linguistics, 4791--4800

    2019

  46. [55]

    Zhang, K.; Ding, N.; Qi, B.; Zhu, X.; Long, X.; and Zhou, B. 2023 a . CR a S h: Clustering, Removing, and Sharing Enhance Fine-tuning without Full Large Language Model. In Conference on Empirical Methods in Natural Language Processing, 9612--9637

    2023

  47. [56]

    V.; et al

    Zhang, S.; Roller, S.; Goyal, N.; Artetxe, M.; Chen, M.; Chen, S.; Dewan, C.; Diab, M.; Li, X.; Lin, X. V.; et al. 2023 b . Opt: Open pre-trained transformer language models. arXiv preprint arXiv:2205.01068

    2023 arXiv

  48. [57]

    C.; and Liu, Z

    Zhou, K.; Yang, J.; Loy, C. C.; and Liu, Z. 2022. Conditional prompt learning for vision-language models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 16816--16825

    2022

  49. [58]

    Z.; and Fredrikson, M

    Zou, A.; Wang, Z.; Kolter, J. Z.; and Fredrikson, M. 2023. Universal and transferable adversarial attacks on aligned language models. arXiv preprint arXiv:2307.15043

    2023 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.