Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

RepLoRA: Reparameterizing Low-Rank Adaptation via the Perspective of Mixture of Experts

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 ↗ Review JSON

Pith's one-line read The paper claims that reparameterizing LoRA's low-rank matrices changes the data needed for a target estimation error from exponential to polynomial, and proposes RepLoRA to realize that gain.

desk verdict A useful empirical PEFT method, but the exponential-to-polynomial theoretical claim does not follow from the paper's own lower bound. read the letter →

arxiv 2502.03044 v2 pith:TWJQPGEX submitted 2025-02-05 cs.LG

Tuan Truong , Chau Nguyen , Huy Nguyen , Minh Le , Trung Le , Nhat Ho This is my paper · ORCID
classification cs.LG
keywords Low-RankAdaptationParameter-EfficientFine-TuningMixtureofExpertsReparameterizationSampleComplexityLoRA
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 tries to establish that the reason LoRA struggles with small datasets is structural: when LoRA's low-rank updates are viewed through the mixture-of-experts (MoE) form of an attention head, estimating the unshared low-rank matrices has minimax convergence slower than any polynomial rate, so reaching error $\epsilon$ can require exponentially many samples. It then argues that reparameterizing the low-rank matrices as outputs of shared linear or nonlinear maps removes this bottleneck, giving rates $O_P(\sqrt{\log(n)/n})$ and polynomial sample complexity. On that basis it proposes RepLoRA, which generates the query and value low-rank matrices from two lightweight MLPs with shared diagonal inputs, and reports that RepLoRA consistently beats vanilla LoRA across language, image, video, and multimodal benchmarks, matching LoRA's full-data accuracy with about 30% of the training data.

What carries the argument

The central object is the MoE reading of an attention head: each output position is a softmax mixture over token-position experts $f_j(X)=W_V^\top x_j$ with score functions $s_{i,j}(X)=x_i^\top W_Q W_K^\top x_j/\sqrt{d_v}$, and LoRA's low-rank updates perturb both. The convergence analysis is carried out with a Voronoi loss that partitions fitted mixture atoms by nearest ground-truth atom; the loss is what turns the least-squares rate into a statement about the low-rank matrices themselves. The mechanism proposed to improve the rate is shared-structure reparameterization: instead of learning $A_Q,A_V,B_Q,B_V$ separately, generate them as outputs of shared maps from common inputs, which the theory shows changes the estimation rate from exponential-scale to polynomial-scale data requirements.

What would settle it

Fit the same MoE regression with the exact RepLoRA parameterization used in practice (separate MLP heads for queries and values, diagonal shared inputs) and compute or simulate the minimax rate of low-rank matrix estimation; if the rate is no better than exponential, the paper's central theoretical claim does not apply to its own method.

Watch

Extended reading notes

Core claim

Viewing each attention head as a softmax-gated MoE, LoRA's learned updates modify both the experts (through value projections) and the gating score functions (through query projections). In the paper's regression model, vanilla LoRA's parameter estimates are shown to have worst-case convergence slower than any polynomial in $n$, which implies exponential sample complexity; reparameterizing the low-rank matrices as outputs of shared maps $A_Q=A_V=W_1A$, $B_Q=B_V=W_2B$, or their nonlinear versions, yields $D_2(\bar G_n,\bar G^*)=O_P(\sqrt{\log(n)/n})$, hence a polynomial number of samples suffices. The paper takes this as evidence that shared-structure reparameterization is the mechanism behind RepLoRA's improved sample efficiency, and the experimental section shows the resulting method improving on LoRA consistently, with the largest gains at small data fractions.

Load-bearing premise

The theoretical polynomial-rate result assumes the query and value low-rank matrices are exactly shared ($A_Q=A_V$ and $B_Q=B_V$), which the method deployed in experiments does not do; if that simplification is not representative, the proven sample-complexity gain may not explain RepLoRA's empirical advantage.

Editorial extensions

If this is right

  • If the theory is right, shared-structure reparameterization is a statistical fix rather than an optimization trick: it removes a minimax inefficiency inherent to estimating LoRA matrices independently.
  • RepLoRA should keep its advantage in low-data regimes; on FGVC it reports gains up to 40.4% relative to LoRA at $f=1\%$ training data and matches LoRA's full-data accuracy at $f=30\%$.
  • Nonlinear reparameterization should dominate linear reparameterization, consistent with the rate improvement of Theorem 5.3 over Theorem 5.2.
  • Because the MLP is discarded after training, the sample-efficiency gain costs no extra inference overhead.
  • The LoRA-as-MoE formulation applies beyond queries and values, so the same reparameterization argument extends naturally to keys and to other low-rank PEFT updates expressed in this form.

Reading between the lines

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

  • A testable extension is to apply the same shared-input MLP reparameterization to other low-rank PEFT variants; if the theory is mechanistic, they should inherit part of the sample-efficiency gain.
  • The paper's strongest empirical claim is about data fraction, but the theory addresses estimation error of low-rank matrices rather than downstream task accuracy; connecting the two would require a task-level transfer argument the paper does not make.
  • The diagonal shared inputs used in practice are a stronger structural assumption than the theory's full shared matrices, so ablating diagonal versus dense shared inputs would reveal whether parameter efficiency or the shared structure itself drives the gain.
  • The lower bound for vanilla LoRA is minimax, so real optimizers with constrained initialization might see a smaller exponential-to-polynomial gap than the worst-case regression model predicts.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 paper proposes RepLoRA, a parameter-efficient fine-tuning method that reparameterizes LoRA's low-rank matrices as outputs of lightweight MLPs with shared inputs. The authors first cast LoRA in a mixture-of-experts framework, then analyze a simplified regression model to argue that reparameterization improves sample complexity from exponential to polynomial. Theorems 5.2 and 5.3 give polynomial upper bounds for the reparameterized estimators, while Theorem 5.1 gives a minimax lower bound for the un-reparameterized estimator. The paper reports extensive experiments on language, vision, video, and multimodal tasks, claiming consistent gains over LoRA and matching LoRA's full-data performance with 30% of the training data.

Significance. If the exponential-to-polynomial claim were established, the paper would provide a substantial theoretical justification for a simple and widely applicable PEFT modification, and the empirical results across four domains would make the contribution practically relevant. The paper has clear strengths: the MoE-based reformulation of LoRA is instructive, the upper-bound analyses in Theorems 5.2 and 5.3 appear internally coherent, and the experimental section is extensive, including a sample-efficiency prediction that is tested directly. However, the central theoretical claim as stated is not supported by the provided results, and the theory is developed under an assumption that the deployed method does not satisfy.

major comments (3)
  1. [Section 5.1, Theorem 5.1] Theorem 5.1 establishes sup_G E[D_{1,r}(Ĝ_n, G)] ≳ n^{-1/2}, which is a polynomial lower bound. The subsequent paragraph's inference that the low-rank estimators are 'slower than any polynomial rates' and may require O(exp(ε^{-1/τ})) samples does not follow: a lower bound of n^{-1/2} is compatible with a parameter-error rate of n^{-1/(2r)} and with a D_{1,r} rate of n^{-1/2}, both polynomial. Moreover, the construction in Appendix A.1 itself yields D_{1,r}(G_n, G*) = O(n^{-r}) with ∥f_{G_n} - f_{G*}∥/D_{1,r} → 0, demonstrating that polynomial loss rates are consistent with the paper's own singularity analysis. The central exponential-to-polynomial sample-complexity claim in the abstract and Section 5 therefore is not proven.
  2. [Section 5.2, Section 6] The reparameterized theory assumes A_Q = A_V and B_Q = B_V (Section 5.2, first paragraph, and the 'without loss of generalization' remark), while the deployed RepLoRA in Section 6, Eq. (18), uses separate output heads for A_Q/A_V and B_Q/B_V with diagonal inputs A and B. The theoretical upper bounds in Theorems 5.2 and 5.3 therefore apply to a different estimator than the one evaluated in Section 7, so the claim that the theory explains RepLoRA's empirical sample-efficiency gains is not supported. The paper should either extend the analysis to the separate-head architecture or explicitly present the theory as a stylized model that is not directly predictive of the deployed method.
  3. [Sections 5.1 and 5.2] The comparison between 'without reparameterization' and 'with reparameterization' is made across incomparable loss functions: Theorem 5.1 is a lower bound on D_{1,r}, which accumulates r-th powers of deviations of four matrices, whereas Theorems 5.2 and 5.3 are upper bounds on D_2 and D_3, which measure deviations of the product W_2 B W_1 A or of the pair (W_1 A, W_2 B) in linear or squared terms. The text does not translate the D_{1,r} lower bound into a lower bound on the metrics used for the reparameterized estimators, so the claimed reduction 'from exponential to polynomial' is not a formally established statement about a common estimation error.
minor comments (5)
  1. [Section 5.1] The phrase 'slower than any polynomial rates O_P(n^{-1/2r})' is internally contradictory because n^{-1/2r} is polynomial for every fixed r; the sentence should be rephrased.
  2. [Section 5.2] 'without loss of generalization' should be 'without loss of generality'.
  3. [Equation (11)] In the definition of the regression function, the expression 'M^0_K × R^{d×d}' appears to be missing a symbol or a phrase; please clarify the intended meaning.
  4. [Table 3] The 'A VG' and 'PPT' columns are not visually separated from the dataset columns, making the table hard to read; consider adding clearer column separators or a note explaining that 'A VG' is the average over the nineteen datasets.
  5. [References] The reference 'Langley, P., Crafting papers on machine learning' is listed but not cited in the text.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the theoretical rates are derived from external statistical machinery; self-citations are present but not load-bearing.

full rationale

The paper's central theoretical results are not circular. Theorem 5.1 is a minimax lower bound proved with Le Cam's lemma and a constructed perturbation sequence in Appendix A.1; Theorem 5.2 follows from the parametric L2 rate of Proposition A.1, obtained via van de Geer's empirical process bound, plus a local identifiability argument under the Voronoi loss. No parameter is fitted and then renamed as a prediction: the sample-efficiency experiment on FGVC is an independent empirical check of the theory, and the claimed 30% data-efficiency figure is not encoded in the theorems. The main self-citations (Le et al., 2024; Le et al., 2025) motivate the MoE view and the reparameterization idea, but the paper re-derives the MoE-LoRA mapping in Eqs. (4)-(9) and proves the rates itself in the appendix, so those citations are not load-bearing reductions of the central derivation. The reviewer-noted weaknesses are not circularity: the inference from a polynomial n^{-1/2} lower bound to 'slower than any polynomial rates' and exponential sample complexity is a logical gap, and the theory's AQ=AV, BQ=BV simplification is a scope mismatch with the deployed method's separate heads. Both are correctness or generalization concerns, not cases where a result is equivalent to its inputs by construction.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central theoretical result rests on a softmax-gated MoE data-generating process, compact parameter space, over-specified number of experts, and strong identifiability assumptions on activations. The most fragile entry is the shared-equality assumption AQ=AV and BQ=BV, which the deployed method does not respect. No new physical or mathematical entities are introduced.

free parameters (4)
  • r (LoRA rank) = 8 (vision), 16/32 (LLaMA-7B/13B), 128 (VL-BART)
    Chosen per task in the experiments. The theoretical statements do not depend on a specific value, but the empirical claims do.
  • alpha (LoRA scaling) = 8, 32, 64, 128 depending on task
    Standard hyperparameter, not fitted to data.
  • MLP hidden dimension = 64
    Chosen by hand for the non-linear reparameterization in Appendix C.2.
  • dimensions m and m' of shared matrices A and B = Not specified
    Free architectural parameters in the reparameterization. The paper says they are 'given dimensions' but does not constrain them in the theory.
assumptions (6)
  • domain assumption Observed data are generated exactly by the softmax-gated MoE regression in Eq. (11) with Gaussian noise.
    Used throughout Section 5. If true LoRA fine-tuning does not match this model, the bounds concern a different problem.
  • domain assumption The number of fitted experts L' is larger than the true number L, and the parameter space Theta is compact.
    Required for the least-squares estimators in Eqs. (12), (14), and (16), and for the Voronoi-loss analysis.
  • domain assumption The input distribution mu has bounded support.
    Invoked in the definition of the L2(mu) norm and in the empirical process arguments.
  • ad hoc to paper Activations sigma1 and sigma2 satisfy algebraic independence, uniform Lipschitz, and strong identifiability (Assumptions A.1-A.3).
    Introduced in Appendix A.3 for Theorem 5.3 and not checked for the sigmoid MLP used in RepLoRA.
  • ad hoc to paper Query low-rank matrices equal value low-rank matrices, AQ=AV=phi1(A) and BQ=BV=phi2(B).
    Section 5.2 calls this a simplification and later claims it is without loss of generality, but the deployed RepLoRA uses separate heads for queries and values.
  • domain assumption The first-row, first-head simplification captures the statistical behavior of LoRA.
    Section 5 restricts the analysis to the first row of the first attention head, without an argument that this represents the full LoRA update.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RepLoRA: Reparameterizing Low-Rank Adaptation via the Perspective of Mixture of Experts." pith.science (2026). https://pith.science/paper/TWJQPGEX

@misc{pith2026250203044,
  author       = {Pith},
  title        = {Pith review of: RepLoRA: Reparameterizing Low-Rank Adaptation via the Perspective of Mixture of Experts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TWJQPGEX}},
  note         = {Machine review of arXiv:2502.03044}
}
read the original abstract

Low-rank Adaptation (LoRA) has emerged as a powerful method for fine-tuning large-scale foundation models. Despite its popularity, the theoretical understanding of LoRA has remained limited. This paper presents a theoretical analysis of LoRA by examining its connection to the Mixture of Experts models. Under this framework, we show that simple reparameterizations of the LoRA matrices can notably accelerate the low-rank matrix estimation process. In particular, we prove that reparameterization can reduce the data needed to achieve a desired estimation error from an exponential to a polynomial scale. Motivated by this insight, we propose Reparameterized Low-Rank Adaptation (RepLoRA), which incorporates lightweight MLPs to reparameterize the LoRA matrices. Extensive experiments across multiple domains demonstrate that RepLoRA consistently outperforms vanilla LoRA. Notably, with limited data, RepLoRA surpasses LoRA by a margin of up to 40.0% and achieves LoRA's performance with only 30.0% of the training data, highlighting both the theoretical and empirical robustness of our PEFT method.

Figures

Figures reproduced from arXiv: 2502.03044 by the authors.

Figure 1
Figure 1. Overview of our proposed method RepLoRA, which reparameterizes the low-rank matrices as the output of a lightweight MLP, whose inputs are two diagonal matrices. models for specific downstream tasks can be prohibitively expensive. To address this challenge, several parameter￾efficient fine-tuning (PEFT) methods (Houlsby et al., 2019; Lester et al., 2021; Jia et al., 2022) have emerged, facili￾tating effective adaptat… view at source ↗
Figure 2
Figure 2. Sample Efficiency on FGVC Datasets. RepLoRA not only outperforms LoRA consistently but also achieves LoRA per￾formance on a full dataset with only f = 30% training fraction. scale to a polynomial scale, thereby substantially improv￾ing sample efficiency. Based on these insights, we intro￾duce Reparameterized Low-Rank Adaptation (RepLoRA) - a novel PEFT method reparameterizes low-rank matrices through a lightweight M… view at source ↗
Figure 3
Figure 3. Performance improvements over LoRA. RepLoRA out￾performs LoRA across all domains, with non-linear reparameteri￾zation substantially surpassing its linear counterpart. 7. Experiments Experimental Settings. We conduct extensive experi￾ments across multiple domains to demonstrate the versatil￾ity and effectiveness of RepLoRA in a wide range of tasks. Our evaluation spans four distinct settings: language (com￾monsense r… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visualization of sample efficiency of LoRA and RepLoRA on five FGVC Datasets. AQ = σ A 1 (A), BQ = σ B 1 (B) AV = σ A 2 (A), BV = σ B 2 (B) For the linear reparameterization, σ A 1 , σA 2 , σB 1 , σB 2 were implemented with linear layers without activation. For the non…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Partial Differential Equation Barriers to Identifiability in Infinite Mixture Models

    math.ST 2026-08 conditional novelty 7.0 of 10

    Kernels annihilated by a non-trivial parameter differential or difference-differential operator make the mixing measure in infinite mixtures non-identifiable, with a minimax lower bound ruling out consistent estimation.

Reference graph

Works this paper leans on

14 extracted references · 6 canonical work pages · cited by 1 Pith paper

  1. [1]

    ,0), Bn V,2 = B∗ V,1 − 1 n(A∗ V,1)(1) (1, 0,

    + 1 2nr+1 and exp(cn i ) = exp(c∗ i−1) for any 3 ≤ i ≤ L + 1; • Bn Q,1 = Bn Q,2 = B∗ Q,1 and Bn Q,i = B∗ Q,i−1 for any 3 ≤ i ≤ L + 1; • An Q,1 = An Q,2 = A∗ Q,1 and An Q,i = A∗ Q,i−1 for any 3 ≤ i ≤ L + 1; • Bn V,1 = B∗ V,1 + 1 n(A∗ V,1)(1) (1, 0, . . . ,0), Bn V,2 = B∗ V,1 − 1 n(A∗ V,1)(1) (1, 0, . . . ,0) and Bn V,i = B∗ V,i−1 for any 3 ≤ i ≤ L + 1, • A...

    work page

  2. [10]

    Radford, A., Kim, J

    Curran Associates Inc. Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pp. 8748–8763. PMLR, 2021. Razdaibiedina, A., Mao, Y ., Khabsa, M., Lewis, M., Hou, R.,...

    2021

  3. [13]

    (20) It can be seen that D1,r(Gn, G∗) → 0 as n → ∞

    + 1 nr+1 i · 1 nr = O(n−r). (20) It can be seen that D1,r(Gn, G∗) → 0 as n → ∞. Subsequently, we illustrate that ∥fGn − fG∗ ∥L2(µ)/D1,r(Gn, G∗) → 0. In particular, let us consider the quantity Qn(X) := h LX k=1 exp(X⊤(M 0 Q + B∗ Q,kA∗ Q,k)X + c∗ k) i · [fGn (X) − f ¯G∗ (X)], 14 RepLoRA: Reparameterizing Low-rank Adaptation from the Perspective of Mixture ...

    work page

  4. [14]

    Moreover, we can also verify thatBn(X) = 0, and Cn(X) = O(n−(r+1))

    + 1 nr+1 i exp(X⊤(M 0 Q + B∗ Q,1A∗ Q,1)X)(Bn V,iAn V,i − B∗ V,1A∗ V,1)X = 1 2 h exp(b∗,1) + 1 nr+1 i exp(X⊤(M 0 Q + B∗ Q,1A∗ Q,1)X)[(Bn V,1An V,1 − B∗ V,1A∗ V,1) + (Bn V,2An V,2 − B∗ V,1A∗ V,1)]X = 0, where the last equality occurs as Bn V,1An V,1 − B∗ V,1A∗ V,1 = 1 n e11 and Bn V,2An V,2 − B∗ V,1A∗ V,1 = − 1 n e11 in which e11 denotes the matrix of size ...

    work page 1997

  5. [161]

    emnlp-main.161/

    URL https://aclanthology.org/2020. emnlp-main.161/. Li, L., Lei, J., Gan, Z., Yu, L., Chen, Y .-C., Pillai, R., Cheng, Y ., Zhou, L., Wang, X., Wang, W. Y ., Wang, W. Y ., Berg, T. L., Bansal, M., Liu, J., Wang, L., and Liu, Z. Value: A multi-task benchmark for video-and- language understanding evaluation. In Vanschoren, J. and Yeung, S. (eds.), Proceedin...

    2020

  6. [243]

    emnlp-main.243/

    URL https://aclanthology.org/2021. emnlp-main.243/. Lewis, M., Liu, Y ., Goyal, N., Ghazvininejad, M., Mo- hamed, A., Levy, O., Stoyanov, V ., and Zettlemoyer, L. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and com- prehension. In Jurafsky, D., Chai, J., Schluter, N., and Tetreault, J. (eds.), Proceeding...

    2021

  7. [319]

    emnlp-main.319/

    URL https://aclanthology.org/2023. emnlp-main.319/. Hudson, D. A. and Manning, C. D. Gqa: A new dataset for real-world visual reasoning and compositional question answering, 2019. Jacobs, R. A., Jordan, M. I., Nowlan, S. J., and Hinton, G. E. Adaptive mixtures of local experts. Neural Computation, 3, 1991. Jia, M., Tang, L., Chen, B.-C., Cardie, C., Belon...

    work page arXiv 2023

  8. [353]

    acl-long.353/

    URL https://aclanthology.org/2021. acl-long.353/. Liu, H., Li, C., Wu, Q., and Lee, Y . J. Visual instruction tun- ing. Advances in neural information processing systems, 36, 2024a. Liu, S.-Y ., Wang, C.-Y ., Yin, H., Molchanov, P., Wang, Y .-C. F., Cheng, K.-T., and Chen, M.-H. Dora: weight- decomposed low-rank adaptation. In Proceedings of the 41st Inte...

    work page 2021

Show all 14 references
  1. [421]

    RepLoRA: Reparameterizing Low-Rank Adaptation via the Perspective of Mixture of Experts

    URL https://aclanthology.org/2023. findings-acl.421/. Riquelme, C., Puigcerver, J., Mustafa, B., Neumann, M., Jenatton, R., Susano Pinto, A., Keysers, D., and Houlsby, N. Scaling vision with sparse mixture of experts. In Ranzato, M., Beygelzimer, A., Dauphin, Y ., Liang, P., a...

    2023 arXiv

  2. [703]

    acl-main.703/

    URL https://aclanthology.org/2020. acl-main.703/. Li, J., Li, D., Xiong, C., and Hoi, S. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation, 2022. Li, L., Chen, Y .-C., Cheng, Y ., Gan, Z., Yu, L., and Liu, J. HERO: Hierarc...

    2020

  3. [2020]

    ISBN 978-3-030-58588-4

    Springer-Verlag. ISBN 978-3-030-58588-4. doi: 10.1007/978-3-030-58589-1 27. URL https://doi. org/10.1007/978-3-030-58589-1_27 . Lester, B., Al-Rfou, R., and Constant, N. The power of scale for parameter-efficient prompt tuning. In Moens, M.-F., Huang, X., Specia, L., and Yih, ...

    doi

  4. [2022]

    Hu, Z., Wang, L., Lan, Y ., Xu, W., Lim, E.-P., Bing, L., Xu, X., Poria, S., and Lee, R

    URL http://dblp.uni-trier.de/db/ conf/iclr/iclr2022.html#HuSWALWWC22. Hu, Z., Wang, L., Lan, Y ., Xu, W., Lim, E.-P., Bing, L., Xu, X., Poria, S., and Lee, R. LLM-adapters: An adapter fam- ily for parameter-efficient fine-tuning of large language models. In Bouamor, H., Pino, ...

    2023 doi

  5. [2023]

    doi: 10.18653/v1/2023.emnlp-main.85

    Association for Computational Linguistics. doi: 10.18653/v1/2023.emnlp-main.85. URL https:// aclanthology.org/2023.emnlp-main.85/. Qiu, Z., Liu, W., Feng, H., Xue, Y ., Feng, Y ., Liu, Z., Zhang, D., Weller, A., and Sch ¨olkopf, B. Controlling text-to- image diffusion by ortho...

    2023 doi

  6. [2799]

    PMLR, 2019. 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. In ICLR. OpenReview.net,

    2019

Pith tools

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