Pith. sign in

REVIEW 4 major objections 4 minor 28 references

Come Together, But Not Right Now: A Progressive Strategy to Boost Low-Rank Adaptation

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

Pith's one-line read CoTo claims that progressively turning on LoRA adapters during fine-tuning — most adapters randomly off at first, all on at the end — yields better accuracy, mergeable solutions, and pruning resilience than standard LoRA.

desk verdict Wide empirical survey of a simple progressive LoRA dropout schedule, but the main gains are entangled with higher learning rates and the theory overclaims. read the letter →

arxiv 2506.05713 v2 pith:XNMM5OL5 submitted 2025-06-06 cs.LG

classification cs.LG
keywords CoTolow-rankadaptationprogressivetrainingadapterdropoutlinearmodeconnectivitymodelmergingpruningrobustnessstochasticregularization
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

CoTo is a training schedule, not a new architecture: over the first 75% of LoRA fine-tuning, each layer's adapter is randomly switched off with probability $1 - p(t)$, where $p(t) = \min(1, 4t/3T)$ rises linearly from 0 to 1, and all adapters stay fully active for the final 25%. The paper's diagnosis is that vanilla LoRA locks adapters into 'lazy' minima near their random initialization, with high-layer adapters absorbing most of the gradient, and its claim is that stochastic early deactivation forces balanced layer-wise optimization and broader exploration of the loss landscape. Consequently it reports consistent gains in single-task accuracy across vision, commonsense, and math benchmarks, much flatter interpolation between independently trained solutions, better multi-task merging, stronger pruning robustness, and 7–24% less training time. The reason this matters is that CoTo requires no architectural change, adds no parameters, and applies uniformly to LoRA variants such as DoRA, HiRA, PiSSA, rsLoRA, LoRA+, and LoRA-Pro, so a sympathetic reader sees a plausible drop-in upgrade for parameter-efficient fine-tuning.

What carries the argument

The load-bearing object is the time-dependent activation schedule $p(t) = \min(1, 4t/3T)$ with per-layer indicators $\delta_i \in \{0,1\}$ drawn independently with $\Pr(\delta_i = 1) = p(t)$, gating each adapter's contribution as $\delta_i \mathbf{1} \odot \Delta W_i$; here 'adapter' means all LoRA parameters inside a single Transformer layer. One mechanism does four jobs: it instantiates the subnetwork curriculum whose expected loss Theorem 3.1 shows to upper-bound a binomially weighted sum of subnetwork losses, it connects training to dropout stability and hence to linear mode connectivity, it makes each adapter's marginal contribution measurable through Shapley values computed by the multilinear extension, and it saves compute because inactive adapters are skipped entirely in forward and backward passes.

What would settle it

Re-run the headline 11-task image classification comparison with LoRA and LoRA-CoTo at the same learning rate (2e-4) and same rank; if the accuracy gain disappears or reverses, the schedule is not the cause. A second decisive check is the $\lambda = 0.5$ interpolation test on commonsense reasoning at matched learning rates, since the linear-mode-connectivity claim is the paper's most distinctive prediction.

Watch

Extended reading notes

Core claim

The central claim is that standard LoRA trains adapters as one monolithic update and pays for it: 'lazy-training' dynamics keep the whole set of low-rank updates near its initialization while a few high-layer adapters monopolize the gradient signal, so the converged solution is layer-imbalanced and survives neither weight averaging nor pruning. CoTo replaces this with a curriculum over subnetworks: at step $t$ each adapter is active with probability $p(t) = \min(1, 4t/3T)$, so early training is done by sparse random subsets of adapters, the model must learn to behave well under arbitrary adapter dropout, and by the final 25% training reverts to ordinary LoRA with all adapters on. The paper supports this with Theorem 3.1, showing for convex loss that the expected CoTo objective upper-bounds a binomially weighted sum of subnetwork losses, and with Shapley-value measurements in which vanilla LoRA concentrates 69% of adapter contribution in the top 4 of 12 layers while CoTo flattens deviations to within 8%. The payoff if the claim holds is that a schedule is worth more than a merger: interpolating two independently trained CoTo solutions retains near-full accuracy (79% versus 39% for LoRA at equal mixing), and merging task-specific CoTo adapters improves linear weight fusion from 47.17% to 58.53% average accuracy on LLaMA-2-7B.

Load-bearing premise

The reported gains come from the progressive on-off schedule itself and not from the higher learning rates CoTo is given in the main experiments, where rates run 2–5 times above the vanilla baselines.

Editorial extensions

If this is right

  • On 11 image-classification tasks with ViT-B/16, CoTo raises average accuracy over vanilla LoRA from 82.95% to 83.48%, and similar or larger gains appear for DoRA (83.45% to 83.93%) and HiRA (83.98% to 84.34%).
  • On 8 commonsense-reasoning tasks, CoTo lifts LoRA from 77.61% to 80.63% average accuracy on LLaMA-2-7B and from 80.79% to 85.02% on LLaMA-3-8B.
  • Independently trained CoTo solutions are linearly mode connected: averaging their adapter weights at interpolation ratio $\lambda = 0.5$ keeps about 79% accuracy on commonsense reasoning, where vanilla LoRA drops to about 39%.
  • Multi-task merging improves across strategies: on LLaMA-2-7B, linear weight fusion of CoTo adapters averages 58.53% versus 47.17% without CoTo, and LoRA-LEGO merging rises from 62.21% to 67.19%.
  • CoTo-trained adapters survive pruning better — about 10 points higher accuracy at 50% unstructured sparsity — and training is faster by 7–24% in wall-clock time because deactivated adapters are skipped.

Reading between the lines

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

  • Editorial extension: CoTo is essentially a curriculum that forces each adapter to be locally useful before it is allowed to rely on its neighbors; the same "on later, together later" principle could be applied to other parameter-efficient modules such as prompts, prefixes, or adapter layers, and it invites a scheduling theory that predicts which curriculum shapes help merging most.
  • Editorial extension: the main comparisons run CoTo at 2–5 times the learning rate of the vanilla baselines (e.g., 5e-4 versus 2e-4 on image classification), so the headline gains are partly attributable to learning rate; the matched-rate evidence is confined to the ablation table, and the cleanest causal test is a matched-rate rerun on the headline benchmarks.
  • Editorial extension: because CoTo trains adapters against random subset dropout, the approximated per-adapter Shapley values could serve as a cheap predictor of merging quality — layers with near-zero marginal contribution could be pruned first without waiting for a full merging experiment.
  • Editorial extension: the activation schedule's shape is a free parameter; the paper's own ablations show exponential schedules trade early convergence for even better merging and pruning, so per-task tuning of the $p(t)$ curve is a plausible next lever.
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 / 4 minor

Summary. The paper proposes CoTo, a progressive training schedule for LoRA fine-tuning in which each adapter's activation probability increases linearly from 0 to 1 over the first 75% of training steps and then stays at 1 for the remaining 25%. The authors claim that this schedule improves single-task accuracy, makes independently trained adapters more linearly mode connectable, improves multi-task merging and pruning robustness, and reduces training compute. The paper includes a Jensen-inequality bound (Theorem 3.1) and a Shapley-value analysis of layer-wise adapter contributions, along with experiments on vision classification, commonsense reasoning, mathematical reasoning, language understanding, and image generation.

Significance. If the causal effect of the schedule were established, CoTo would be a simple, broadly applicable, and practically useful addition to LoRA-based fine-tuning: it requires no architectural changes, is compatible with several LoRA variants, and the compute savings are plausible. The empirical program is extensive, with three seeds, multiple backbones, and both structured and unstructured pruning, and the authors ship code. However, the theoretical contribution is modest, and the main experimental comparisons currently conflate the schedule with a higher learning rate, so the primary empirical claim is not yet fully supported.

major comments (4)
  1. [§4.1, Table 8, Table 10] The headline comparisons in Tables 1–4 use different learning rates for the baseline and CoTo runs. Table 8 shows LoRA at 2e-4 and LoRA-CoTo at 5e-4 on all 11 vision tasks; Table 10 reports default versus CoTo learning rates of 1e-5 versus 5e-5 for commonsense reasoning, 2e-5 versus 1e-4 for mathematical reasoning, and 5e-5 versus 2e-4 for image classification. The only matched-learning-rate experiment, Table 6, shows a 0.13 pp gain at 5e-5, a 1.31 pp gain at 1e-4, and at 2e-4 the LoRA-Pro baseline collapses to 40.91 while CoTo maintains 56.25. This pattern suggests that the main gains could arise from CoTo's ability to stabilize or exploit a higher learning rate rather than from the progressive activation schedule itself. The paper should either report headline results with matched learning rates or explicitly frame CoTo as a method that enables higher-LR training and provide evidence that the progressive schedule contributes beyond the LR change.
  2. [§3.4, Theorem 3.1] Theorem 3.1 is a direct application of Jensen's inequality to the Bernoulli mixture; it holds for any dropout distribution over adapters and does not establish that CoTo trains solutions that are robust to adapter deletion or linearly mode connected. The sentence 'Prior work... shows that dropout stability often implies that independently trained solutions can be connected by a low-loss linear path' is a heuristic appeal rather than a proof, and the theorem itself does not mention the progressive schedule p(t). The abstract and Section 3.4 therefore overstate the theoretical support. I suggest restating the theorem as a generic bound on the expected subnetwork loss (which is correct) and presenting the LMC and pruning results as empirical findings only.
  3. [§3.5, §4.4] The Shapley-value analysis uses the loss ℓ as the value function v(R) in Eq. (7). Because a lower loss is better, the marginal contribution v(R ∪ {i}) − v(R) is expected to be negative when adding an adapter improves the model, and the reported 'contributions' in Figure 8 (e.g., 69% concentrated in the top layers) are not defined in terms of this sign. Please clarify whether the figure plots φ_i(v), −φ_i(v), or a normalized quantity, and how the concentration percentages are computed.
  4. [Table 1, Table 9] Several per-task gains in Table 1 are smaller than one standard deviation reported in Table 9 (e.g., Caltech 96.25±0.12 vs 96.26±0.10; ImageNet 73.49±0.06 vs 73.53±0.04). The paper does not test whether the average improvements are statistically significant. A paired test across the 11 tasks or reporting the number of tasks where the improvement exceeds the standard deviation would substantiate the claim of 'consistent' improvement.
minor comments (4)
  1. [Eq. (2)] The linear schedule starts at p(1) = 4/(3T) rather than exactly 0 as stated in the text; if the authors intend p(0) = 0, the formula should be adjusted accordingly, for example by using (t−1)/(3T/4) in the first branch.
  2. [Footnote 3] The citation (Zhao et al., 2024a) for the linear model ensemble appears to point to LoraRetriever, which does not define this operation; please cite the original source of the linear model ensemble formula.
  3. [Figure 2] The weight functions w_j(p) are plotted for L = 6, but the text does not specify the value of L used in the figure; adding this detail would aid readability.
  4. [Table 7] The reported training speedups come from skipping inactive adapters, but the paper does not discuss whether the wall-clock measurements include the overhead of sampling activation indicators or whether the speedup would persist on hardware with specialized kernels; a brief clarification would be useful.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: CoTo's schedule is not fitted to the metrics it predicts, and its gains are benchmark comparisons; the learning-rate mismatch is a correctness risk, not circularity.

full rationale

The derivation chain is self-contained rather than circular. CoTo defines an activation schedule p(t) (Eq. 2) and an objective (Eq. 4); Theorem 3.1 is a direct application of Jensen's inequality to that objective and does not assume the paper's empirical conclusions about accuracy, merging, or pruning. The LMC and Shapley-value discussions are post-hoc analyses or empirical measurements, not fitted parameters renamed as predictions. The headline comparisons use baselines from prior work, including the authors' own HiRA paper (Table 2 note: "Results without CoTo are copied from (Huang et al., 2025)"), but this self-citation supplies external baseline numbers rather than a load-bearing axiom; the central claim does not reduce to it. The main substantive concern is that CoTo is often run at a higher learning rate than the baseline (Tables 8 and 10: e.g., 2e-4 vs 5e-4 on vision, 1e-5 vs 5e-5 on commonsense reasoning), which may confound the measured gains, and the matched-LR ablation in Table 6 shows small gains at the lowest learning rate. That is a threat to causal attribution, not a circularity of the derivation chain. No equation or fitted constant is shown to be equivalent by construction to the reported predictions, so the circularity score is low.

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

CoTo introduces no new physical or model entities; it is a training schedule applied to existing LoRA adapters. The free parameters are the schedule's phase proportion, its shape, and the grid-searched learning rates used in the experiments.

free parameters (3)
  • first-phase proportion = 75%
    The fraction of training steps where p(t) < 1; chosen by hand and ablated in Figure 9, left panel.
  • activation schedule shape = linear
    p(t) = 4t/3T; chosen by hand; Appendix C.1 compares exponential and sine schedules.
  • CoTo learning rate = 2e-4 to 5e-2 depending on task and variant
    Grid-searched separately for CoTo runs in the vision benchmark (Table 8); a potential confound for the reported gains.
assumptions (3)
  • standard math Convexity of the loss ℓ(·, y)
    Theorem 3.1 requires ℓ convex; cross-entropy is convex in logits, but the network output is not convex in adapter weights, so the bound is not tight.
  • domain assumption Layer-wise imbalance causes suboptimal minima
    The motivation that higher layers dominate gradient updates is taken from cited work and not verified in this paper beyond the Shapley analysis.
  • domain assumption Dropout stability implies linear mode connectivity
    Invoked via Frankle et al. and Adilova et al.; the paper does not prove CoTo satisfies the conditions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Come Together, But Not Right Now: A Progressive Strategy to Boost Low-Rank Adaptation." pith.science (2026). https://pith.science/paper/XNMM5OL5

@misc{pith2026250605713,
  author       = {Pith},
  title        = {Pith review of: Come Together, But Not Right Now: A Progressive Strategy to Boost Low-Rank Adaptation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XNMM5OL5}},
  note         = {Machine review of arXiv:2506.05713}
}
read the original abstract

Low-rank adaptation (LoRA) has emerged as a leading parameter-efficient fine-tuning technique for adapting large foundation models, yet it often locks adapters into suboptimal minima near their initialization. This hampers model generalization and limits downstream operators such as adapter merging and pruning. Here, we propose CoTo, a progressive training strategy that gradually increases adapters' activation probability over the course of fine-tuning. By stochastically deactivating adapters, CoTo encourages more balanced optimization and broader exploration of the loss landscape. We provide a theoretical analysis showing that CoTo promotes layer-wise dropout stability and linear mode connectivity, and we adopt a cooperative-game approach to quantify each adapter's marginal contribution. Extensive experiments demonstrate that CoTo consistently boosts single-task performance, enhances multi-task merging accuracy, improves pruning robustness, and reduces training overhead, all while remaining compatible with diverse LoRA variants. Code is available at https://github.com/zwebzone/coto.

Figures

Figures reproduced from arXiv: 2506.05713 by the authors.

Figure 1
Figure 1. Illustration of the CoTo progressive activation schedule for LoRA adapters. For the first 75% of training (i.e., t < 3T /4), each adapter is stochastically deactivated (shown in gray), where the activation probability p(t) increases linearly from 0 to 1 as training progresses. In the final 25% of training, p(t) = 1, and all adapters remain active, reducing to full fine-tuning. downstream tasks. By introducing a smal… view at source ↗
Figure 2
Figure 2. Visualization of the weight function wj (p) in Eq. (6). where 1 is an all-ones matrix of the same size as Wi , and ⊙ denotes the Hadamard product. The training objective is to minimize the expected loss: min {∆Wi} Eδ [ℓ(yˆ, y)] , (4) where δ = [δ1, . . . , δL] ⊺ ∈ {0, 1} L, y is the target label, and ℓ is typically cross-entropy loss for classification or mean squared error for regression. 3.3. Training Dynamics Cur… view at source ↗
Figure 3
Figure 3. Linear interpolation accuracy on commonsense reason￾ing tasks (Hu et al., 2023). CoTo’s interpolation curve averaged across 8 tasks (orange) remains flatter and higher compared to vanilla LoRA (blue), demonstrating superior LMC. lazy convergence, especially as model capacity increases. We further evaluate CoTo on mathematical reasoning tasks by fine-tuning LLaMA-2-7B on MetaMathQA (Yu et al., 2024) and testing it on… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Linear interpolation accuracy on 11 image classification tasks. CoTo (orange) consistently outperforms vanilla LoRA (blue). Adding alignment (denoted by † ) improves both but preserves CoTo’s margin. ture), LoRA-CoTo maintains 79% accuracy, whereas vanilla LoRA drops t…
Figure 5
Figure 5. Figure 5: Analysis of the optimized alignment matrix P for LoRA† and LoRA-CoTo† . Error bars denote standard deviations. modest. Similar trends are observed for LLaMA-2-13B, indicating that CoTo encourages each adapter to converge to parameters that lie in closer, low-loss subsp…
Figure 7
Figure 7. Figure 7: Average accuracy (%) on model pruning for the vi￾sual texture classification task (Cimpoi et al., 2014). Left panel: Structured pruning applied to LoRA, LoRA-CoTo, and Early LoRA-CoTo under four pruning patterns: alternating layers (EveryOther), first 4 layers (Low), m…
Figure 6
Figure 6. Figure 6: Customized sample images generated by SDXL (Podell et al., 2024) with and without CoTo. When merging style and ob￾ject adapters via ZipLoRA (Shah et al., 2025), CoTo preserves both the object identity and artistic style more faithfully than vanilla LoRA. Each compariso…
Figure 8
Figure 8. Figure 8: Approximated Shapley values of LoRA adapters by mul￾tilinear extension (Owen, 1972). 0 1 16 1 4 1 2 3 4 1 0.54 0.55 0.56 0.57 0.58 0.90 0.92 0.94 0.96 0.98 1.00 Accuracy Time Base CoTo-L CoTo CoTo-H 54 56 58 Accuracy [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Ablation analysis of CoTo. Left panel: Impact of varying the proportion of training time allocated to the first (stochastic activation) phase on model accuracy (purple) and normalized train￾ing time (orange). Right panel: Comparison of dropout strategies, including no …
Figure 10
Figure 10. Figure 10: Average accuracy (%) on model merging and pruning for the visual texture classification task (Cimpoi et al., 2014). Left panel: Merging accuracy when λ = 0.5. Right panel: Pruning accuracy when removing alternating layers (i.e., EveryOther). adapters and undermines ba…
Figure 11
Figure 11. Figure 11: Linear interpolation accuracy on 8 individual commonsense reasoning tasks (Hu et al., 2023). All EveryOther Low Middle High Query Key Value 30 40 50 Aircraft All EveryOther Low Middle High Query Key Value 94 95 96 97 Caltech All EveryOther Low Middle High Query Key Va…
Figure 12
Figure 12. Figure 12: Average accuracy (%) on structured model pruning for 11 image classification tasks. B.3. Multi-Task Merging for Discriminative Language Understanding CoTo improves merging accuracy by 4.18%–6.18% across linear weight fusion, linear model ensemble, and LoRA￾LEGO (Zhao …
Figure 13
Figure 13. Figure 13: Performance evolution during training under different activation schedules on the visual texture classification task (Cimpoi et al., 2014). Merging accuracy is measured at λ = 0.5. Pruning accuracy is measured for EveryOther (i.e., removing alternating layers). 40 20 …
Figure 14
Figure 14. Figure 14: t-SNE visualization (Van der Maaten & Hinton, 2008) of the learned weights of LoRA adapters across five seeds and four learning rates, under three perplexity settings. Black dots denote initialization points, and color gradients indicate different learning rates. D. P…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 10 canonical work pages

  1. [5]

    For image classification, we follow Zanella & Ben Ayed (2024) and use 11 datasets: • Aircraft (Maji et al.,

    image generation. For image classification, we follow Zanella & Ben Ayed (2024) and use 11 datasets: • Aircraft (Maji et al.,

  2. [10]

    Fine-grained visual classification of aircraft

    Maji, S., Rahtu, E., Kannala, J., Blaschko, M., and Vedaldi, A. Fine-grained visual classification of aircraft. arXiv preprint arXiv:1306.5151,

  3. [11]

    PiSSA: Principal sin- gular values and singular vectors adaptation of large lan- guage models

    Meng, F., Wang, Z., and Zhang, M. PiSSA: Principal sin- gular values and singular vectors adaptation of large lan- guage models. arXiv preprint arXiv:2404.02948,

  4. [13]

    The Llama 3 herd of models

    Meta. The Llama 3 herd of models. arXiv preprint arXiv:2407.21783,

  5. [14]

    SocialIQA: Commonsense reasoning about social interactions

    Sap, M., Rashkin, H., Chen, D., LeBras, R., and Choi, Y . SocialIQA: Commonsense reasoning about social interactions. arXiv preprint arXiv:1904.09728,

  6. [15]

    R., and Shah, M

    Soomro, K., Zamir, A. R., and Shah, M. UCF101: A dataset of 101 human actions classes from videos in the wild. arXiv preprint arXiv:1212.0402,

  7. [17]

    LoRA meets dropout under a unified framework

    Wang, S., Chen, L., Jiang, J., Xue, B., Kong, L., and Wu, C. LoRA meets dropout under a unified framework. arXiv preprint arXiv:2403.00812, 2024a. Wang, S., Yu, L., and Li, J. LoRA-GA: Low-rank adaptation with gradient approximation. In Advances in Neural Information Processing Systems, 2024b. Wang, Z., Liang, J., He, R., Wang, Z., and Tan, T. LoRA- Pro: ...

  8. [18]

    HellaSwag: Can a machine really finish your sentence? arXiv preprint arXiv:1905.07830,

    Zellers, R., Holtzman, A., Bisk, Y ., Farhadi, A., and Choi, Y . HellaSwag: Can a machine really finish your sentence? arXiv preprint arXiv:1905.07830,

Show all 28 references
  1. [19]

    LoRA-FA: Memory-efficient low-rank adaptation for large language models fine-tuning

    Zhang, L., Zhang, L., Shi, S., Chu, X., and Li, B. LoRA-FA: Memory-efficient low-rank adaptation for large language models fine-tuning. arXiv preprint arXiv:2308.03303 , 2023a. Zhang, Q., Chen, M., Bukharin, A., He, P., Cheng, Y ., Chen, W., and Zhao, T. Adaptive budget alloca...

  2. [21]

    (human action classification) For commonsense reasoning, we use 8 tasks from Commonsense170K (Hu et al., 2023): • ARC-c and ARC-e (Clark et al.,

  3. [22]

    For language understanding, we follow Zhao et al

    and test on GSM8K (Cobbe et al., 2021). For language understanding, we follow Zhao et al. (2024b) and use 9 tasks from GLUE (Wang et al.,

  4. [23]

    and Flan Collection (Longpre et al., 2023): • CoLA (Dolan & Brockett,

  5. [25]

    backbone. Learning Rate Method CoLA MRPC QNLI QQP RTE SST2 Avg w/o CoTo 5e-4 LoRA 87.44 89.46 94.31 91.05 85.56 95.18 90.50 Fusion 69.89 68.38 49.97 65.50 47.29 55.05 59.35 Ensemble 69.13 31.62 50.54 63.31 52.71 50.92 53.04 LoRA-LEGO 73.28 33.15 74.19 80.95 61.46 69.18 65.37 1...

  6. [26]

    Linear interpolation accuracy on 8 individual commonsense reasoning tasks (Hu et al., 2023). All EveryOther Low Middle High Query Key Value 30 40 50 Aircraft All EveryOther Low Middle High Query Key Value 94 95 96 97 Caltech All EveryOther Low Middle High Query Key Value 70 75...

  7. [27]

    lazy training

    confirm LoRA adapters converge near initialization (i.e., “lazy training”) across learning rates. CoTo yields tighter clusters under initialization noise. 17 CoTo: A Progressive Strategy to Boost Low-Rank Adaptation 0 2000 4000 6000 8000 0.0 0.2 0.4 0.6 0.8 1.0 Activation Prob...

  8. [28]

    Merging accuracy is measured at λ = 0.5

    Performance evolution during training under different activation schedules on the visual texture classification task (Cimpoi et al., 2014). Merging accuracy is measured at λ = 0.5. Pruning accuracy is measured for EveryOther (i.e., removing alternating layers). 40 20 0 20 40 x...

  9. [2005]

    S., Zhai, X., Poczos, B., and Singh, A

    Du, S. S., Zhai, X., Poczos, B., and Singh, A. Gradient descent provably optimizes over-parameterized neural networks. arXiv preprint arXiv:1810.02054,

  10. [2012]

    Dropout: a simple way to prevent neural networks from overfitting

    Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., and Salakhutdinov, R. Dropout: a simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1):1929–1958,

  11. [2013]

    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. arXiv preprint arXiv:2104.08691,

  12. [2014]

    R., Angeli, G., Potts, C., and Manning, C

    Bowman, S. R., Angeli, G., Potts, C., and Manning, C. D. A large annotated corpus for learning natural language inference. arXiv preprint arXiv:1508.05326,

  13. [2016]

    (question-answering) • QQP 5 (Quora questions) • RTE (Dagan et al., 2005; Bar Haim et al., 2006; Giampiccolo et al., 2007; Bentivogli et al.,

  14. [2018]

    Training verifiers to solve math word problems

    Cobbe, K., Kosaraju, V ., Bavarian, M., Chen, M., Jun, H., Kaiser, L., Plappert, M., Tworek, J., Hilton, J., Nakano, R., et al. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168,

  15. [2019]

    Think you have solved question answering? Try ARC, the AI2 reasoning chal- lenge

    Clark, P., Cowhey, I., Etzioni, O., Khot, T., Sabharwal, A., Schoenick, C., and Tafjord, O. Think you have solved question answering? Try ARC, the AI2 reasoning chal- lenge. arXiv preprint arXiv:1803.05457,

  16. [2020]

    Higher layers need more LoRA experts

    Gao, C., Chen, K., Rao, J., Sun, B., Liu, R., Peng, D., Zhang, Y ., Guo, X., Yang, J., and Subrahmanian, V . Higher layers need more LoRA experts. arXiv preprint arXiv:2402.08562, 2024a. Gao, Z., Wang, Q., Chen, A., Liu, Z., Wu, B., Chen, L., and Li, J. Parameter-efficient fin...

  17. [2021]

    LoRA dropout as a sparsity regularizer for over- fitting control

    Lin, Y ., Ma, X., Chu, X., Jin, Y ., Yang, Z., Wang, Y ., and Mei, H. LoRA dropout as a sparsity regularizer for over- fitting control. arXiv preprint arXiv:2404.09610,

  18. [2023]

    Y ., Pang, T., Du, C., and Lin, M

    Huang, C., Liu, Q., Lin, B. Y ., Pang, T., Du, C., and Lin, M. LoraHub: Efficient cross-task generalization via dynamic LoRA composition. arXiv preprint arXiv:2307.13269,

  19. [2024]

    Llama 2: Open foundation and fine-tuned chat models

    Meta. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288,

  20. [2025]

    A rank stabilization scaling factor for fine- tuning with LoRA

    Kalajdzievski, D. A rank stabilization scaling factor for fine- tuning with LoRA. arXiv preprint arXiv:2312.03732 ,

Pith tools

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