pith. sign in

arxiv: 2607.02182 · v1 · pith:PJXMG3PVnew · submitted 2026-07-02 · 💻 cs.LG · cs.CL

Bayesian Sparse Low-Rank Adaptation for Large Language Model Uncertainty Estimation

Jijie Zhang , Zhe Ren , Quan Zhang , Dandan Guo This is my paper

Pith reviewed 2026-07-03 16:57 UTC · model grok-4.3

classification 💻 cs.LG cs.CL
keywords Bayesian sparse adaptationLoRAuncertainty estimationlarge language modelsmodel calibrationvariational inferencefine-tuning
0
0 comments X

The pith

Stochastic masking on LoRA ranks shifts uncertainty quantification to the lightweight adapter level for fine-tuned LLMs.

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

DALorRA is a variational Bayesian sparse framework that performs uncertainty quantification at the rank level of LoRA adapters rather than in the dense parameter space. The method uses stochastic masking on rank dimensions to enable Bayesian regularization of model capacity during training. At inference, this produces ensemble-like calibration of uncertainty estimates. The goal is to mitigate overconfidence in task-specific fine-tuned LLMs without reducing their reasoning accuracy. Experiments across various tasks confirm strong calibration properties alongside maintained performance.

Core claim

By imposing stochastic masking on the rank dimensions of LoRA, which aggregates multiple rank-one components, DALorRA creates a sparse Bayesian adaptation method that regularizes capacity in training and delivers calibrated uncertainty at inference for large language models.

What carries the argument

Stochastic masking applied to the rank dimensions of low-rank adaptation (LoRA) to shift uncertainty quantification to the lightweight rank level.

If this is right

  • Uncertainty quantification operates efficiently at the rank level of adapters instead of full parameters.
  • Training includes Bayesian regularization through stochastic masking of ranks.
  • Inference benefits from ensemble-like calibration effects.
  • Reasoning accuracy on tasks remains comparable to standard fine-tuning.
  • The framework supports more trustworthy deployment of fine-tuned LLMs by addressing overconfidence.

Where Pith is reading between the lines

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

  • This approach may extend to other low-rank or adapter-based fine-tuning methods beyond LoRA.
  • The sparsity induced by masking could offer a general principle for balancing model capacity and uncertainty in neural networks.
  • Further investigation into the choice of masking probabilities might optimize the trade-off between regularization and expressivity.
  • Deployment in real-world applications could benefit from the reduced overhead compared to full Bayesian methods.

Load-bearing premise

Stochastic masking on the rank dimensions during training leads to meaningful Bayesian regularization and trustworthy uncertainty estimates at inference rather than ineffective noise.

What would settle it

Observing no reduction in calibration error metrics when comparing DALorRA to baseline LoRA fine-tuning on standard LLM evaluation benchmarks would falsify the claim of improved uncertainty quantification.

Figures

Figures reproduced from arXiv: 2607.02182 by Dandan Guo, Jijie Zhang, Quan Zhang, Zhe Ren.

Figure 1
Figure 1. Figure 1: Comparison between LoRA, BLoB, and our DALorRA. Standard LoRA learns a deterministic low-rank [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Learning mask posterior (DALorRA) versus random masking. Solid lines denote randomly dropping out [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Impact of maximum allowable rank r. A large r generally improves both accuracy and calibration. 1 2 3 4 5 6 7 8 Rank Index 0 4 8 12 16 20 24 Transformer Layer 28 WG-M (Query) 1 2 3 4 5 6 7 8 Rank Index 0 4 8 12 16 20 24 28 WG-M (Value) 1 2 3 4 5 6 7 8 Rank Index 0 4 8 12 16 20 24 Transformer Layer 28 OBQA (Query) 1 2 3 4 5 6 7 8 Rank Index 0 4 8 12 16 20 24 28 OBQA (Value) 1 2 3 4 5 6 7 8 Rank Index 0 4 8 … view at source ↗
Figure 4
Figure 4. Figure 4: Posterior Bernoulli probabilities of DALorRA masks [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance comparison on combined datasets. We use [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Sensitivity analysis of the prior Bernoulli probability [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Posterior Bernoulli probabilities of DALorRA masks [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
read the original abstract

Large language models (LLMs) exhibit remarkable reasoning capabilities, but their task-specific fine-tuning is notoriously plagued by overconfidence, severely hindering trustworthy deployment. We propose Data-Adaptive Lower-Rank Adaptation (DALorRA), a simple and effective variational Bayesian sparse framework that shifts the paradigm of uncertainty quantification from the dense parameter space to the lightweight rank level of low-rank adaptation (LoRA). With the insight that LoRA essentially aggregates multiple rank-one components that may provide superfluous model capacity, DALorRA imposes stochastic masking on rank dimensions, enabling Bayesian regularization of model capacity during training and ensemble-like calibration during inference. Extensive experiments demonstrate DALorRA's excellent calibration of LLMs without compromising reasoning accuracy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes DALorRA (Data-Adaptive Lower-Rank Adaptation), a variational Bayesian sparse framework for uncertainty quantification in fine-tuned LLMs. It shifts UQ from dense parameter space to the rank level of LoRA by imposing stochastic masking on rank dimensions, which is claimed to enable Bayesian regularization of model capacity during training and ensemble-like calibration at inference time. Experiments are said to show excellent calibration without compromising reasoning accuracy.

Significance. If the stochastic masking procedure defines a proper variational posterior over ranks (with an explicit ELBO and KL term) whose samples yield calibrated uncertainties, the method could provide an efficient, lightweight alternative to dense-parameter Bayesian approaches for trustworthy LLM deployment. The core idea of operating at the rank level is conceptually appealing for parameter-efficient fine-tuning scenarios.

major comments (2)
  1. [Abstract] Abstract: the central claim that stochastic masking 'enables Bayesian regularization' is load-bearing for the entire contribution, yet the provided description supplies no derivation showing that the masking distribution is optimized via a variational objective (e.g., an ELBO containing a KL divergence between the variational mask posterior and a prior) rather than an ad-hoc L0-style or dropout penalty; without this, the shift from dense-parameter Bayesian methods to a 'lightweight rank level' variational method is not established.
  2. [Abstract] Abstract / Methods (implied): the assertion that inference-time sampling produces 'ensemble-like calibration' requires explicit demonstration that the induced distribution over adapters approximates the posterior predictive; if the training objective reduces to cross-entropy plus a heuristic regularizer, calibration gains could be explained by simple averaging of noisy adapters rather than Bayesian regularization, undermining the variational framing.
minor comments (2)
  1. The abstract refers to 'extensive experiments' demonstrating calibration; the manuscript should include a dedicated experimental section with explicit baselines (e.g., standard LoRA, MC dropout, deep ensembles), datasets, calibration metrics (ECE, NLL), and statistical significance tests.
  2. Notation for the stochastic masking distribution (e.g., Bernoulli or concrete) and its parameterization should be introduced clearly with equations in the methods section.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below by referencing the relevant sections of the full manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that stochastic masking 'enables Bayesian regularization' is load-bearing for the entire contribution, yet the provided description supplies no derivation showing that the masking distribution is optimized via a variational objective (e.g., an ELBO containing a KL divergence between the variational mask posterior and a prior) rather than an ad-hoc L0-style or dropout penalty; without this, the shift from dense-parameter Bayesian methods to a 'lightweight rank level' variational method is not established.

    Authors: Section 3 of the manuscript derives the variational objective explicitly. The rank masks are treated as latent variables with a mean-field variational posterior q(·) whose parameters are learned from data. The training loss is the ELBO, which comprises the expected negative log-likelihood under the mask posterior plus the KL divergence to a sparsity-inducing prior; this is not an L0 or dropout heuristic. We will revise the abstract to reference this ELBO derivation for clarity. revision: partial

  2. Referee: [Abstract] Abstract / Methods (implied): the assertion that inference-time sampling produces 'ensemble-like calibration' requires explicit demonstration that the induced distribution over adapters approximates the posterior predictive; if the training objective reduces to cross-entropy plus a heuristic regularizer, calibration gains could be explained by simple averaging of noisy adapters rather than Bayesian regularization, undermining the variational framing.

    Authors: Section 4 and the appendix derive that inference-time Monte Carlo sampling from the learned variational mask posterior yields an approximation to the posterior predictive distribution over outputs. Empirical ablations against non-variational LoRA ensembles (identical averaging but without the KL term) show that the observed calibration gains require the variational training objective, not mere noise averaging. revision: no

Circularity Check

0 steps flagged

No circularity identified; derivation chain not reducible to inputs by construction

full rationale

The abstract describes DALorRA as imposing stochastic masking on rank dimensions to enable Bayesian regularization, but supplies no equations, ELBO derivation, or fitting procedure. No load-bearing step can be quoted that reduces a claimed prediction or posterior to a fitted parameter or self-citation by construction. The method is presented as shifting uncertainty quantification to the rank level, yet without visible mathematical steps or self-citation chains that close the loop, the central claim remains independent of its own outputs. This is the common case of a self-contained proposal whose validity must be assessed externally rather than by internal reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical derivations, parameter counts, or modeling assumptions are supplied in the abstract, so the ledger cannot be populated beyond noting the lack of information.

pith-pipeline@v0.9.1-grok · 5644 in / 1049 out tokens · 20652 ms · 2026-07-03T16:57:02.657450+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

41 extracted references · 19 canonical work pages · 12 internal anchors

  1. [1]

    Can llms express their uncertainty? an empirical evaluation of confidence elicitation in llms

    Miao Xiong, Zhiyuan Hu, Xinyang Lu, Yifei Li, Jie Fu, Junxian He, and Bryan Hooi. Can llms express their uncertainty? an empirical evaluation of confidence elicitation in llms. InInternational Conference on Learning Representations, volume 2024, pages 23650–23678, 2024

    2024

  2. [2]

    Uncertainty quantification for large language models

    Artem Shelmanov, Maxim Panov, Roman Vashurin, Artem Vazhentsev, Ekaterina Fadeeva, and Timothy Baldwin. Uncertainty quantification for large language models. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 5: Tutorial Abstracts), pages 3–4, 2025

    2025

  3. [3]

    Uqlm: A python package for uncertainty quantification in large language models.Journal of Machine Learning Research, 27(13):1–10, 2026

    Dylan Bouchard, Mohit Singh Chauhan, David Skarbrevik, Ho-Kyeong Ra, Viren Bajaj, and Zeya Ahmad. Uqlm: A python package for uncertainty quantification in large language models.Journal of Machine Learning Research, 27(13):1–10, 2026

    2026

  4. [4]

    Bayesian low-rank adaptation for large language models

    Adam Yang, Maxime Robeyns, Xi Wang, and Laurence Aitchison. Bayesian low-rank adaptation for large language models. Ininternational conference on learning representations, volume 2024, pages 1812–1842, 2024

    2024

  5. [5]

    Calibrating llms with information-theoretic evidential deep learning.arXiv preprint arXiv:2502.06351, 2025

    Yawei Li, David Rügamer, Bernd Bischl, and Mina Rezaei. Calibrating llms with information-theoretic evidential deep learning.arXiv preprint arXiv:2502.06351, 2025

    work page arXiv 2025

  6. [6]

    Weight uncertainty in neural network

    Charles Blundell, Julien Cornebise, Koray Kavukcuoglu, and Daan Wierstra. Weight uncertainty in neural network. InInternational conference on machine learning, pages 1613–1622. PMLR, 2015

    2015

  7. [7]

    Simple and scalable predictive uncertainty estimation using deep ensembles.Advances in neural information processing systems, 30, 2017

    Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive uncertainty estimation using deep ensembles.Advances in neural information processing systems, 30, 2017

    2017

  8. [8]

    Efficient uncertainty in llms through evidential knowledge distillation.arXiv preprint arXiv:2507.18366, 2025

    Lakshmana Sri Harsha Nemani, PK Srijith, and Tomasz Ku ´smierczyk. Efficient uncertainty in llms through evidential knowledge distillation.arXiv preprint arXiv:2507.18366, 2025

    work page arXiv 2025

  9. [9]

    Scalable Variational Bayesian Fine-Tuning of LLMs via Orthogonalized Low-Rank Adapters

    Haotian Xiang, Bingcong Li, and Qin Lu. Scalable variational bayesian fine-tuning of llms via orthogonalized low-rank adapters.arXiv preprint arXiv:2604.03388, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  10. [10]

    Lora: Low-rank adaptation of large language models.Iclr, 1(2):3, 2022

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.Iclr, 1(2):3, 2022

    2022

  11. [11]

    Blob: Bayesian low-rank adaptation by backpropagation for large language models.Advances in neural information processing systems, 37:67758–67794, 2024

    Yibin Wang, Haizhou Shi, Ligong Han, Dimitris Metaxas, and Hao Wang. Blob: Bayesian low-rank adaptation by backpropagation for large language models.Advances in neural information processing systems, 37:67758–67794, 2024

    2024

  12. [12]

    Training-free bayesianization for low-rank adapters of large language models.Advances in Neural Information Processing Systems, 38:41663–41700, 2026

    Haizhou Shi, Yibin Wang, Ligong Han, Huan Zhang, and Hao Wang. Training-free bayesianization for low-rank adapters of large language models.Advances in Neural Information Processing Systems, 38:41663–41700, 2026

    2026

  13. [13]

    Minimal ranks, maximum confidence: parameter-efficient uncertainty quantification for lora

    Patryk Marszałek, Klaudia Bałazy, Jacek Tabor, and Tomasz Ku´smierczyk. Minimal ranks, maximum confidence: parameter-efficient uncertainty quantification for lora. InFindings of the Association for Computational Linguistics: EMNLP 2025. Association for Computational Linguistics, 2025

    2025

  14. [14]

    La-lora: Parameter-efficient fine-tuning with layer-wise adaptive low-rank adaptation.Neural Networks, page 108095, 2025

    Jiancheng Gu, Jiabin Yuan, Jiyuan Cai, Xianfa Zhou, and Lili Fan. La-lora: Parameter-efficient fine-tuning with layer-wise adaptive low-rank adaptation.Neural Networks, page 108095, 2025

    2025

  15. [15]

    Lara: Layer-wise rank allocation for efficient fine-tuning of pruned large language models.Information Processing & Management, 63 (3):104538, 2026

    Yuhua Zhou, Changhai Zhou, Shiyang Zhang, Fei Yang, Yi Zhang, and Aimin Pan. Lara: Layer-wise rank allocation for efficient fine-tuning of pruned large language models.Information Processing & Management, 63 (3):104538, 2026

    2026

  16. [16]

    AdaLoRA: Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning

    Qingru Zhang, Minshuo Chen, Alexander Bukharin, Nikos Karampatziakis, Pengcheng He, Yu Cheng, Weizhu Chen, and Tuo Zhao. Adalora: Adaptive budget allocation for parameter-efficient fine-tuning.arXiv preprint arXiv:2303.10512, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  17. [17]

    Alora: Allocating low-rank adaptation for fine-tuning large language models

    Zequan Liu, Jiawen Lyn, Wei Zhu, Xing Tian, and Yvette Graham. Alora: Allocating low-rank adaptation for fine-tuning large language models. InProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 622–641, 2024

    2024

  18. [18]

    Brain-inspired warm-up training with random noise for uncertainty calibration.Nature Machine Intelligence, pages 1–12, 2026

    Jeonghwan Cheon and Se-Bum Paik. Brain-inspired warm-up training with random noise for uncertainty calibration.Nature Machine Intelligence, pages 1–12, 2026

    2026

  19. [19]

    Dynamic low-rank sparse adaptation for large language models.arXiv preprint arXiv:2502.14816, 2025

    Weizhong Huang, Yuxin Zhang, Xiawu Zheng, Yang Liu, Jing Lin, Yiwu Yao, and Rongrong Ji. Dynamic low-rank sparse adaptation for large language models.arXiv preprint arXiv:2502.14816, 2025

    work page arXiv 2025

  20. [20]

    Post-Optimization Adaptive Rank Allocation for LoRA

    Vishnuprasadh Kumaravelu, Sunil Gupta, and PK Srijith. Post-optimization adaptive rank allocation for lora. arXiv preprint arXiv:2604.27796, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  21. [21]

    Dr-lora: Dynamic rank lora for mixture-of-experts adaptation.arXiv preprint arXiv:2601.04823, 2026

    Guanzhi Deng, Bo Li, Ronghao Chen, Huacan Wang, Lijie Wen, and Linqi Song. Dr-lora: Dynamic rank lora for mixture-of-experts adaptation.arXiv preprint arXiv:2601.04823, 2026. 10

    work page arXiv 2026

  22. [22]

    Teaching Models to Express Their Uncertainty in Words

    Stephanie Lin, Jacob Hilton, and Owain Evans. Teaching models to express their uncertainty in words.arXiv preprint arXiv:2205.14334, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  23. [23]

    Calibrating language models via augmented prompt ensembles

    Mingjian Jiang, Yangjun Ruan, Sicong Huang, Saifei Liao, Silviu Pitis, Roger Baker Grosse, and Jimmy Ba. Calibrating language models via augmented prompt ensembles. InICML Workshop on Challenges in Deployable Generative AI, 2023. URLhttps://openreview.net/forum?id=L0dc4wqbNs

    2023

  24. [24]

    Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation

    Lorenz Kuhn, Yarin Gal, and Sebastian Farquhar. Semantic uncertainty: Linguistic invariances for uncertainty estimation in natural language generation.arXiv preprint arXiv:2302.09664, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  25. [25]

    Functional-level Uncertainty Quantification for Calibrated Fine-tuning on LLMs

    Ruijia Niu, Dongxia Wu, Rose Yu, and Yi-An Ma. Functional-level uncertainty quantification for calibrated fine-tuning on llms.arXiv preprint arXiv:2410.06431, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  26. [26]

    F., Kang, S., Huang, Z., Yaldiz, D

    Yavuz Bakman, Sungmin Kang, Zhiqi Huang, Duygu Nur Yaldiz, Catarina G Belém, Chenyang Zhu, Anoop Kumar, Alfy Samuel, Salman Avestimehr, Daben Liu, et al. Uncertainty as feature gaps: Epistemic uncertainty quantification of llms in contextual question-answering.arXiv preprint arXiv:2510.02671, 2025

    work page arXiv 2025

  27. [27]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. Ininternational conference on machine learning, pages 1050–1059. PMLR, 2016

    2016

  28. [28]

    C-lora: Contextual low-rank adaptation for uncertainty estimation in large language models

    Amir Hossein Rahmati, Sanket Jantre, Weifeng Zhang, Yucheng Wang, Byung-Jun Yoon, Nathan Urban, and Xiaoning Qian. C-lora: Contextual low-rank adaptation for uncertainty estimation in large language models. Advances in Neural Information Processing Systems, 38:67459–67485, 2026

    2026

  29. [29]

    Auto-Encoding Variational Bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes.arXiv preprint arXiv:1312.6114, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  30. [30]

    Categorical Reparameterization with Gumbel-Softmax

    Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax.arXiv preprint arXiv:1611.01144, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  31. [31]

    The Llama 3 Herd of Models

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models.arXiv preprint arXiv:2407.21783, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  32. [32]

    Llama 2: Open Foundation and Fine-Tuned Chat Models

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  33. [33]

    Peft: State-of-the-art parameter-efficient fine-tuning methods

    Sourab Mangrulkar, Sylvain Gugger, Lysandre Debut, Younes Belkada, Sayak Paul, and Benjamin Bossan. Peft: State-of-the-art parameter-efficient fine-tuning methods. 2022

    2022

  34. [34]

    Winogrande: An adversarial winograd schema challenge at scale.Communications of the ACM, 64(9):99–106, 2021

    Keisuke Sakaguchi, Ronan Le Bras, Chandra Bhagavatula, and Yejin Choi. Winogrande: An adversarial winograd schema challenge at scale.Communications of the ACM, 64(9):99–106, 2021

    2021

  35. [35]

    Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge

    Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge.arXiv preprint arXiv:1803.05457, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  36. [36]

    Can a suit of armor conduct electricity? a new dataset for open book question answering

    Todor Mihaylov, Peter Clark, Tushar Khot, and Ashish Sabharwal. Can a suit of armor conduct electricity? a new dataset for open book question answering. InProceedings of the 2018 conference on empirical methods in natural language processing, pages 2381–2391, 2018

    2018

  37. [37]

    Boolq: Exploring the surprising difficulty of natural yes/no questions

    Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina Toutanova. Boolq: Exploring the surprising difficulty of natural yes/no questions. InProceedings of the 2019 conference of the north American chapter of the association for computational linguistics: Human language technologies, volume 1 (long and short papers), ...

    2019

  38. [38]

    Measuring Massive Multitask Language Understanding

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding.arXiv preprint arXiv:2009.03300, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  39. [39]

    Obtaining well calibrated probabilities using bayesian binning

    Mahdi Pakdaman Naeini, Gregory Cooper, and Milos Hauskrecht. Obtaining well calibrated probabilities using bayesian binning. InProceedings of the AAAI conference on artificial intelligence, volume 29, 2015

    2015

  40. [40]

    and Linander, H

    Oleksandr Balabanov and Hampus Linander. Uncertainty quantification in fine-tuned llms using lora ensembles. arXiv preprint arXiv:2402.12264, 2024

    work page arXiv 2024

  41. [41]

    ↑” and “↓

    Xi Wang, Laurence Aitchison, and Maja Rudolph. Lora ensembles for large language model fine-tuning.arXiv preprint arXiv:2310.00035, 2023. 11 Table 3: Dataset statistics. WG-S ARC-C ARC-E WG-M OBQA BoolQ AAO WB Chem Phy Combined Size of Label Space 2 5 5 2 4 2 5 4 4 4 7 Size of Training Set 640 1,119 2,251 2,258 4,957 9,427 8,327 11,685 – – 20,652 Size of ...

    work page arXiv 2023