Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Anatomy of Unlearning: The Dual Impact of Fact Salience and Model Fine-Tuning

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read This paper establishes that machine unlearning is jointly shaped by fact popularity and the model's training stage, with supervised-fine-tuned models forgetting smoothly and retaining 10-50% more knowledge than pretrained models.

desk verdict The DUET benchmark is a genuine resource, but the paper's central claim about pretrained vs SFT unlearning is compromised because the SFT model was trained on the exact evaluation pairs. read the letter →

arxiv 2602.19612 v5 pith:5TZU32T3 submitted 2026-02-23 cs.CL

Anna Borisiuk , Andrey Savchenko , Alexander Panchenko , Elena Tutubalina This is my paper · ORCID
classification cs.CL
keywords machineunlearningfactpopularitysupervisedfine-tuningpretrainedmodelsknowledgeretentionLLMbenchmarkingWikidatacatastrophicforgetting
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 argues that the standard assumption that all facts are equally forgettable is wrong. It introduces DUET, a benchmark of 28.6k Wikidata-derived question-answer pairs annotated with fact popularity, and shows that the outcome of unlearning depends on both how popular a fact is and whether the model is a pretrained base model or one that has gone through supervised fine-tuning (SFT). The central experimental finding is that SFT models forget more smoothly, tune more stably, and retain 10-50% more of the retain set, while pretrained models are unstable, prone to catastrophic forgetting, and can even improve on popular facts when unlearning is applied. If this result holds, unlearning pipelines and benchmarks must stratify by fact popularity and account for the training stage, and a preliminary SFT step becomes a practical way to stabilize forgetting.

What carries the argument

The central object is the DUET benchmark: 28.6k Wikidata-derived question-answer triplets, each annotated with a popularity score based on Wikipedia sitelink counts and validated against LLM-based salience judgments. The benchmark stratifies facts into rare and popular forget sets at 1%, 5%, and 10% scales, with a retain intersection and compact retain subsets for efficient evaluation. The paper then compares Pretrained and SFT versions of LLaMA-3.1-8B (and Gemma-7B, Qwen-2.5 7B) under three gradient-based unlearning algorithms. The popularity annotation is what carries the argument: it turns unlearning from a single scalar outcome into a function of where the fact sits in the model's memory

What would settle it

Train an SFT model on DUET with the forget and retain evaluation sets held out (or replaced by a disjoint set of comparable facts), re-run the unlearning comparison, and see whether the retention advantage and smoother forgetting persist. If they disappear, the paper's headline claim is confounded.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that pretrained and supervised-fine-tuned variants of the same architecture respond qualitatively differently to the same unlearning algorithms. On popular facts, the pretrained model's ROUGE-L on the forget set increases during gradient ascent, gradient difference, and NPO, as if the unlearning signal were additional fine-tuning, whereas the SFT model forgets as intended. On the retain set, SFT models show 10-50% higher retention and roughly half the risk of catastrophic collapse, and token-level and hidden-state diagnostics show that SFT produces localized representation changes while pretrained models barely change internally. The paper concludes

Load-bearing premise

The claim that supervised fine-tuning stabilizes unlearning rests on comparing an SFT model that was trained on the exact forget/retain evaluation examples against a pretrained model that never saw them, so part or all of the observed advantage may be double exposure to the test items rather than a property of the training stage.

Editorial extensions

If this is right

  • Unlearning evaluations will need to report results stratified by fact popularity; aggregate curves hide the fact that popular and rare facts erase at different rates.
  • A preliminary supervised fine-tuning step on the forget data becomes a practical recipe for more stable, controllable forgetting with higher retention.
  • Pretrained models can 'relearn' popular facts during unlearning, so practitioners who skip fine-tuning risk the exact opposite of the intended effect.
  • SFT models' retention risk is concentrated in popular-fact removal, while pretrained models degrade abruptly regardless of fact type, pointing to different failure modes.

Reading between the lines

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

  • The paper attributes the SFT advantage to the training stage, but the SFT model was fine-tuned on the same evaluation examples; a holdout-SFT control would test whether the 10-50% retention gain is a training-regime effect or a double-exposure artifact.
  • Wikipedia link popularity is a plausible proxy for pretraining corpus frequency; a direct test would measure how unlearning difficulty tracks token counts in the base model's pretraining data.
  • The rare/popular asymmetry suggests adaptive unlearning recipes: stronger signals for rare facts, gentler ones for popular facts, with an optional preliminary SFT pass.
  • If the double-exposure confound is real, the paper's main quantitative claim would shrink, but the qualitative popularity asymmetry is independently supported by the multi-domain and intrinsic-analysis results in the appendix.
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 / 5 minor

Summary. The paper introduces DUET, a 28.6k Wikidata-derived benchmark of question–answer pairs annotated with fact popularity via Wikipedia sitelink counts and LLM-based salience scores. Using this benchmark, the authors compare machine unlearning on a pretrained LLaMA-3.1-8B checkpoint and an SFT variant trained with LoRA on the full DUET set, applying GA, GD, and NPO across rare/popular forget splits. The central empirical claim is that pretrained and SFT models respond qualitatively differently to unlearning: SFT models exhibit smoother forgetting and 10–50% higher retention, while pretrained models are unstable and can even 'relearn' popular facts. The same qualitative pattern is reported for Gemma-7B, Qwen-2.5-7B, an additional algorithm (UNDIAL), and an LLM-as-a-Judge protocol.

Significance. If established, the claim that unlearning behavior depends jointly on fact popularity and training stage would be an important contribution to LLM unlearning evaluation. The DUET benchmark itself, with its popularity stratification and multi-domain subsets, is a potentially useful resource, and the paper works with multiple models, multiple unlearning algorithms, and several evaluation lenses. The main evidence, however, does not currently support the headline causal claim. The SFT model was trained on the same DUET data that is later split into the forget and retain evaluation sets, so the pretrained-vs-SFT contrast conflates training stage with direct exposure to the exact evaluation examples. The paper also lacks error bars, reports a single configuration as the main result, and its own Table 1 contradicts the unqualified '10–50% higher retention' statement. The benchmark and the research question are valuable, but the central comparison needs a cleaner experimental design before the conclusions can be accepted.

major comments (4)
  1. [§3.1, Figure 2] The main confound is that the SFT variant is trained on the full DUET dataset (28.6k samples) while the forget and retain evaluation sets are subsets of DUET. The SFT model therefore receives additional gradient steps on the exact question–answer pairs used to measure forgetting and retention, whereas the pretrained checkpoint has not seen any of them. The abstract's own phrasing — 'An SFT step on the forget data' — makes this exposure part of the treatment. The observed 10–50% higher retention and smoother forgetting may simply reflect double exposure to the evaluation items, not a general property of supervised fine-tuning. This confound is not mentioned in the Limitations section. The central causal claim is not identifiable from the current design. To fix it, the SFT training data must exclude the forget and retain evaluation splits (or the SFT model must be compared against a pretra
  2. [Table 1] The abstract and Section 3.2 claim '10–50% higher retention' for SFT models, but Table 1 shows the opposite on popular facts. For popular forget at lr=2e-5, SFT retain ROUGE-L is lower than pretrained retain ROUGE-L for all three algorithms: GA 0.733 vs 0.864, GD 0.760 vs 0.884, NPO 0.849 vs 0.888. The claim is only supported on rare facts, and even there the size varies by algorithm (e.g., NPO rare: 0.944 vs 0.790). The unqualified '10–50%' statement is therefore not supported by the paper's main table. This is a load-bearing numerical claim and must be corrected, qualified by popularity, or accompanied by the full breakdown.
  3. [§3.1, Figure 3] No uncertainty quantification is reported. The main results appear to come from a single run at a single learning rate, with no seeds, no error bars, and no statistical tests. The hyperparameters (two epochs, lr=2e-5) are selected post hoc from the same curves shown in Figure 3, so the reported differences may be partly overfit to the chosen configuration. Since the paper makes quantitative claims (e.g., '10–50% higher retention', 'roughly half the risk of catastrophic forgetting'), it should report multiple random seeds/initiations and show variability, or at minimum explicitly state that these are single-run observations.
  4. [Abstract vs §3.1] The abstract says 'An SFT step on the forget data yields smoother forgetting...' but Section 3.1 describes the SFT variant as trained on the full DUET dataset, not specifically on the forget set. These are different treatments. If the intended comparison is SFT-on-forget-data, the experiments do not implement it. If the intended comparison is SFT-on-all-data, the abstract should be reworded, and the exposure confound in the first major comment still applies. The paper should clarify exactly which SFT training set is used and align the treatment description throughout.
minor comments (5)
  1. [Figure 1 caption] The figure is a 2x2 grid with small text; consider enlarging or using a table format to improve readability.
  2. [Figure 3/4/5 captions] The captions say 'Left: forget; right: retain' but do not specify axes or the exact split names in a consistent way. Please state the x-axis (learning rate) and y-axis (ROUGE-L) explicitly.
  3. [Limitations] The Limitations section lists scope, popularity proxies, and metrics, but it does not acknowledge the training/evaluation overlap that is the main threat to validity. Please address it directly.
  4. [Section C.6, Table 9] The UNDIAL results table lacks a column for the algorithm name in each block; it appears only in the section text. Consider adding an 'Algorithm' column for self-contained readability.
  5. [Section C.4, Table 6] The table reports token-level rank changes but not standard deviations. Since ranks can be highly variable across examples, adding dispersion or a short note would help assess reliability.

Circularity Check

2 steps flagged · score 7.0 of 10

SFT variant is trained on the full DUET dataset, whose forget/retain subsets are later used for evaluation; the pretrained-vs-SFT contrast conflates training stage with double exposure to the exact test items.

  1. self definitional [Section 3.1 (Experimental Setup) with Section 2 (Popularity-Based Splits and City Sets)]
    "we use the released checkpoint as the Pretrained model and train an SFT variant on the full DUET dataset (28.6k samples) using LoRA"

    Section 2 defines the rare/popular forget sets and the retain intersection as subsets of DUET, and the city forget/retain sets are drawn from the same DUET places-city domain. Therefore the SFT variant was trained on the exact QA pairs later used as forget and retain evaluation sets, while the Pretrained checkpoint was not. The reported higher retention and smoother forgetting of SFT thus conflate the training-stage variable with double exposure to the evaluation items; the outcome is partly guaranteed by the training-set construction rather than being an independent effect of SFT.

  2. fitted input called prediction [Abstract (also Conclusion)]
    "An SFT step on the forget data yields smoother forgetting, more stable tuning, and 10-50% higher retention, while direct unlearning on pretrained models remains unstable and prone to relearning or catastrophic forgetting."

    The 'forget data' for the SFT step is the full DUET dataset, which includes the forget and retain subsets used for evaluation. The headline '10-50% higher retention' is measured on retain pairs that the SFT model has already been fine-tuned on, so the claim is not an independent generalization about training stage; it is an expected consequence of having optimized on those very examples. The paper does not control for this exposure, and the Limitations section does not list it.

full rationale

The central comparison is not self-contained as a causal claim about training stage. Section 3.1 trains the SFT variant on the full DUET dataset; Section 2 builds the rare/popular forget sets and retain sets as subsets of DUET, and the city fast-retain set comes from the same places-city subset. Therefore the SFT model has seen the exact question–answer pairs used to measure forgetting and retention, while the pretrained checkpoint has not. The paper's headline ('SFT step on the forget data yields ... 10-50% higher retention') is thus partly an artifact of double exposure: higher retention on the retain set is expected because those examples were in the SFT training set, and smoother forgetting on the forget set is easier because the model was fine-tuned on those examples before unlearning. The Limitations section lists scope, popularity proxies, and metrics but does not mention this exposure confound. This is the load-bearing circular step; it invalidates the attribution of the observed differences to the training stage. No self-citation chain or uniqueness theorem is load-bearing: the citations to Dontsov et al. and Huang et al. are methodological and not used to force the central result. The paper does include external MMLU/HellaSwag checks, but those test general capability preservation, not the central retention comparison. Additionally, Table 1 itself contradicts the unqualified retention claim (e.g., SFT GradAscent popular retain 0.733 vs pretrain 0.864), underscoring that the comparison is not controlled. Given that the central quantitative claim reduces in part to training-set memorization, a score of 7 is appropriate; it is not a full 10 because the unlearning dynamics after the SFT step are still empirically measured and not strictly tautological.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The quantitative conclusions rely on hand-chosen thresholds and hyperparameters, and the SFT-vs-pretrained comparison is confounded by training on the evaluation set.

free parameters (3)
  • BERT cosine filtering thresholds = 0.6 (main), 0.7 (city retain)
    Used to decide which Wikidata triplets the model 'knows'; changes benchmark composition and no sensitivity analysis is reported.
  • Unlearning learning rate and epoch count = lr=2e-5, epochs=2
    Selected by grid search as a 'practical compromise' after inspecting results; headline numbers depend on this choice.
  • Retain set size = 500 samples from Places City
    Retention claims are based on a 500-sample subset of one domain; not shown to be representative of the full benchmark.
assumptions (4)
  • domain assumption Wikidata triplets with unique answers and BERT cosine >0.6 are facts the model knows.
    Used to construct DUET; filtering is model-specific and threshold hand-chosen.
  • domain assumption Wikipedia sitelink counts are a valid proxy for model-internal fact salience.
    Validated only against LLaMA-3.3-70B judgments on two subsets; not across models or languages.
  • domain assumption ROUGE-L decrease on the forget set measures forgetting, not rephrasing.
    Main tables use ROUGE-L; LLM-judge validation is relegated to the appendix and not used for the headline claims.
  • ad hoc to paper A LoRA fine-tune on the DUET dataset is a representative SFT model.
    The SFT model is trained on the same data later used for evaluation, conflating training stage with double exposure.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Anatomy of Unlearning: The Dual Impact of Fact Salience and Model Fine-Tuning." pith.science (2026). https://pith.science/paper/5TZU32T3

@misc{pith2026260219612,
  author       = {Pith},
  title        = {Pith review of: Anatomy of Unlearning: The Dual Impact of Fact Salience and Model Fine-Tuning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5TZU32T3}},
  note         = {Machine review of arXiv:2602.19612}
}
read the original abstract

Machine Unlearning (MU) enables Large Language Models (LLMs) to remove unsafe or outdated information. However, existing work assumes that all facts are equally forgettable and largely ignores whether the forgotten knowledge originates from pretraining or supervised fine-tuning (SFT). In this paper, we introduce DUET (Dual Unlearning Evaluation across Training Stages), a benchmark of 28.6k Wikidata-derived triplets annotated with fact popularity using Wikipedia link counts and LLM-based salience scores. Our experiments show that pretrained and SFT models respond differently to unlearning. An SFT step on the forget data yields smoother forgetting, more stable tuning, and 10-50% higher retention, while direct unlearning on pretrained models remains unstable and prone to relearning or catastrophic forgetting.

Figures

Figures reproduced from arXiv: 2602.19612 by the authors.

Figure 1
Figure 1. Unlearning landscape across fact popularity [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the DUET benchmark. (a) Topic distribution across 25 semantic themes (28.6k validated QA pairs), dominated by the places city domain. (b) The filtering and stratification to derive rare and popular forget sets and the retain intersection.(c) Compact retain subsets (retain 500 and retain 1500) drawn from places city, providing resource-efficient yet structurally consistent evaluation settings. els respond… view at source ↗
Figure 3
Figure 3. City, n = 5%. Top: rare; bottom: popular. Left: forget (city_forget_{rare,popular}_5); right: retain (city_fast_retain_500). ROUGE is reported across learning rates. Pretrain is shown as a solid line with circles; SFT as a dashed line with stars. Baselines are solid horizontal lines. full DUET dataset (28.6k samples) using LoRA instead of full fine-tuning for efficiency and stabil￾ity. Indeed, parameter-efficient me… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: City, n = 1%. Top: rare; bottom: popular. Left: forget (city_forget_{rare,popular}_1); right: retain (city_fast_retain_500). ROUGE is reported across learning rates. Pretrain is shown as a solid line with circles; SFT as a dashed line with stars. Baselines are solid ho…
Figure 5
Figure 5. Figure 5: City, n = 10%. Top: rare; bottom: popular. Left: forget (city_forget_{rare,popular}_1); right: retain (city_fast_retain_500). ROUGE is reported across learning rates. Pretrain is shown as a solid line with circles; SFT as a dashed line with stars. Baselines are solid h…
Figure 6
Figure 6. Figure 6: Distributions of popularity scores (pop_sum) for the city domain in the new and old benchmark splits. The left panel shows the rare forget 5% subset and the right panel the popular forget 5% subset. Out of 482 samples, 87 differ from the previous version, resulting in …
Figure 7
Figure 7. Figure 7: City split (n = 5%) evaluated on Gemma 7B. Top: rare; bottom: popular. Left: forget sets (city_forget_{rare,popular}_5); right: retain set (city_fast_retain_500). ROUGE scores are shown across learning rates. Pretrained models are plotted as solid lines with circles, a…
Figure 8
Figure 8. Figure 8: City split (n = 5%) evaluated on Qwen-2.5 7B. Top: rare; bottom: popular. Left: forget sets (city_forget_{rare,popular}_5); right: retain set (city_fast_retain_500). ROUGE scores are shown across learning rates. Pretrained models are plotted as solid lines with circles…
Figure 9
Figure 9. Figure 9: Evaluation on LLM benchmarks. Left: MMLU accuracy. Right: HellaSwag accuracy (acc_norm). Curves [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Review history (2 revisions) →

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. Multilingual Unlearning in LLMs: Transfer, Dynamics, and Reversibility

    cs.CL 2026-06 unverdicted novelty 6.0 of 10

    Unlearning in multilingual LLMs suppresses rather than erases knowledge in later layers, with transfer varying by language similarity and reversible via inference-time steering.

Reference graph

Works this paper leans on

26 extracted references · 1 canonical work pages · cited by 1 Pith paper

  1. [1]

    Yinzhi Cao and Junfeng Yang. 2015. Towards making systems forget with machine unlearning. In 2015 IEEE symposium on security and privacy, pages 463--480. IEEE

    2015

  2. [2]

    Ning Ding, Yujia Qin, Guang Yang, Fuchao Wei, Zonghan Yang, Yusheng Su, Shengding Hu, Yulin Chen, Chi-Min Chan, Weize Chen, and 1 others. 2023. Parameter-efficient fine-tuning of large-scale pre-trained language models. Nature machine intelligence, 5(3):220--235

    2023

  3. [3]

    Yijiang River Dong, Hongzhou Lin, Mikhail Belkin, Ramon Huerta, and Ivan Vuli \'c . 2025. Undial: Self-distillation with adjusted logits for robust unlearning in large language models. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long P...

    2025

  4. [4]

    Alexey Dontsov, Dmitrii Korzh, Alexey Zhavoronkin, Boris Mikheev, Denis Bobkov, Aibek Alanov, Oleg Rogov, Ivan Oseledets, and Elena Tutubalina. 2025. https://doi.org/10.18653/v1/2025.findings-acl.1058 CLEAR : Character unlearning in textual and visual modalities . In Findings of the Association for Computational Linguistics: ACL 2025, pages 20582--20603, ...

    work page doi:10.18653/v1/2025.findings-acl.1058 2025

  5. [5]

    Jiahui Geng, Qing Li, Herbert Woisetschlaeger, Zongxiong Chen, Fengyu Cai, Yuxia Wang, Preslav Nakov, Hans-Arno Jacobsen, and Fakhri Karray. 2025. A comprehensive survey of machine unlearning techniques for large language models. arXiv preprint arXiv:2503.01854

    arXiv 2025

  6. [6]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, and 1 others. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations

  7. [7]

    Shengyuan Hu, Yiwei Fu, Zhiwei Steven Wu, and Virginia Smith. 2024. Unlearning or obfuscating? jogging the memory of unlearned llms via benign relearning. arXiv preprint arXiv:2406.13356

    arXiv 2024

  8. [8]

    Baixiang Huang, Canyu Chen, Xiongxiao Xu, Ali Payani, and Kai Shu. 2025. https://openreview.net/forum?id=hmDt068MoZ Can knowledge editing really correct hallucinations? In The Thirteenth International Conference on Learning Representations

    2025

Show all 26 references
  1. [9]

    Joel Jang, Dongkeun Yoon, Sohee Yang, Sungmin Cha, Moontae Lee, Lajanugen Logeswaran, and Minjoon Seo. 2022. Knowledge unlearning for mitigating privacy risks in language models. arXiv preprint arXiv:2210.01504

    2022 arXiv

  2. [10]

    Meghdad Kurmanji, Peter Triantafillou, Jamie Hayes, and Eleni Triantafillou. 2023. Towards unbounded machine unlearning. Advances in neural information processing systems, 36:1957--1987

    2023

  3. [11]

    Nathaniel Li, Alexander Pan, Anjali Gopal, Summer Yue, Daniel Berrios, Alice Gatti, Justin D Li, Ann-Kathrin Dombrowski, Shashwat Goel, Long Phan, and 1 others. 2024. The wmdp benchmark: Measuring and reducing malicious use with unlearning. arXiv preprint arXiv:2403.03218

    2024 arXiv

  4. [12]

    Bo Liu, Qiang Liu, and Peter Stone. 2022. Continual learning and private unlearning. In Conference on Lifelong Learning Agents, pages 243--254. PMLR

    2022

  5. [13]

    Lipton, and J

    Pratyush Maini, Zhili Feng, Avi Schwarzschild, Zachary C. Lipton, and J. Zico Kolter. 2024 a . Tofu: A task of fictitious unlearning for llms

    2024

  6. [14]

    Pratyush Maini, Zhili Feng, Avi Schwarzschild, Zachary C Lipton, and J Zico Kolter. 2024 b . Tofu: A task of fictitious unlearning for llms. arXiv preprint arXiv:2401.06121

    2024 arXiv

  7. [15]

    Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. When not to trust language models: Investigating effectiveness of parametric and non-parametric memories. In Proceedings of the 61st Annual Meeting of the Association for Compu...

    2023

  8. [16]

    Alessandro Mantelero. 2013. The eu proposal for a general data protection regulation and the roots of the â right to be forgottenâ. Computer Law & Security Review, 29(3):229--235

    2013

  9. [17]

    Ayush Sekhari, Jayadev Acharya, Gautam Kamath, and Ananda Theertha Suresh. 2021. Remember what you want to forget: Algorithms for machine unlearning. Advances in Neural Information Processing Systems, 34:18075--18086

    2021

  10. [18]

    Weijia Shi, Jaechan Lee, Yangsibo Huang, Sadhika Malladi, Jieyu Zhao, Ari Holtzman, Daogao Liu, Luke Zettlemoyer, Noah A Smith, and Chiyuan Zhang. 2024. Muse: Machine unlearning six-way evaluation for language models. CoRR

    2024

  11. [19]

    Nianwen Si, Hao Zhang, Heyu Chang, Wenlin Zhang, Dan Qu, and Weiqiang Zhang. 2023. Knowledge unlearning for llms: Tasks, methods, and challenges. arXiv preprint arXiv:2311.15766

    2023 arXiv

  12. [20]

    Xianghui Sun, Yunjie Ji, Baochang Ma, and Xiangang Li. 2023. A comparative study between full-parameter and lora-based fine-tuning on chinese instruction data for instruction following large language model. arXiv preprint arXiv:2304.08109

    2023 arXiv

  13. [21]

    Lingzhi Wang, Xingshan Zeng, Jinsong Guo, Kam-Fai Wong, and Georg Gottlob. 2025. Selective forgetting: Advancing machine unlearning techniques and evaluation in language models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 843--851

    2025

  14. [22]

    Xiaoyu Xu, Xiang Yue, Yang Liu, Qingqing Ye, Haibo Hu, and Minxin Du. 2025. Unlearning isn't deletion: Investigating reversibility of machine unlearning in llms. arXiv preprint arXiv:2505.16831

    2025 arXiv

  15. [23]

    Hongbang Yuan, Zhuoran Jin, Pengfei Cao, Yubo Chen, Kang Liu, and Jun Zhao. 2025. Towards robust knowledge unlearning: An adversarial framework for assessing and improving unlearning robustness in large language models. In Proceedings of the AAAI Conference on Artificial Intel...

    2025

  16. [24]

    Ruiqi Zhang, Licong Lin, Yu Bai, and Song Mei. 2024. Negative preference optimization: From catastrophic collapse to effective unlearning. arXiv preprint arXiv:2404.05868

    2024 arXiv

  17. [25]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  18. [26]

    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 gl...

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.