Pith. sign in

REVIEW 4 major objections 5 minor 44 references

Model Merging for Knowledge Editing

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that early-stopped fine-tuning on a single transformer layer followed by scaled, pruned merging with the base model lets a 7B LLM absorb thousands of factual edits at near-99% success while keeping general capabilities…

desk verdict A sensible batch fine-tune-and-merge recipe is oversold as a sequential-editing method, with internally inconsistent numbers that undercut the experiments. read the letter →

arxiv 2506.12384 v1 pith:UQJNFBM5 submitted 2025-06-14 cs.AI cs.CLcs.LG

Zichuan Fu , Xian Wu , Guojing Li , Yingying Zhang , Yefeng Zheng , Tianshi Ming , Yejing Wang , Wanyu Wang
show 1 more author
Xiangyu Zhao
This is my paper
classification cs.AIcs.CLcs.LG
keywords knowledgeeditingmodelmergingtaskvectorsupervisedfine-tuningsequentialearlystoppinglargelanguagemodels
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 show that knowledge editing need not choose between updating facts and preserving a language model's general abilities. It proposes a two-stage recipe: first fine-tune on all target edits at once, updating only the feed-forward networks of a single transformer layer with sample-wise early stopping; then merge the fine-tuned model back toward the original base by scaling the parameter difference and pruning it to the largest changes. On Qwen2.5-7B the recipe reaches edit success above 96% on ZsRE and most other KnowEdit subsets while C-Eval accuracy stays within about a point of the base model. The paper reads this as evidence that the approach outperforms existing editing methods in sequential editing and preserves general capabilities without architectural changes.

What carries the argument

Two mechanisms carry the argument. First, R-SFT: sample-wise consecutive gradient updates with an early-stop loss threshold, applied only to the feed-forward networks of the fifth transformer layer, so each edit is memorized without broad drift. Second, the merging identity: the task vector $\Delta\theta=\theta_{\text{sft}}-\theta_{\text{base}}$ is scaled by $(1-\alpha)$ and pruned to its top-$p\%$ parameter changes, turning fine-tuning into a sparse additive correction to the base model. The scaling/pruning step is what recovers general capability after fine-tuning degrades it.

What would settle it

Run the framework in a genuinely sequential protocol: take an ordered list of a few hundred facts, apply R-SFT and merging incrementally on slices (or one fact at a time), and after each step re-evaluate edit success on every previously edited fact; if retention of early facts falls materially below the reported ~96% success, or if ordering changes final performance, the sequential-editing claim fails.

Watch

Extended reading notes

Core claim

The central claim is that robust supervised fine-tuning followed by model merging solves the edit-versus-preserve trade-off: the edited model is constructed as $\theta_{\text{edited}} = \theta_{\text{base}} + (1-\alpha)\,\mathrm{Top}_p(\theta_{\text{sft}}-\theta_{\text{base}})$, where $\theta_{\text{sft}}$ comes from sample-wise early-stopped supervised fine-tuning restricted to the feed-forward networks of a single transformer layer. The paper reports edit success of 96.95% on ZsRE, with generalization of 91.58%, while C-Eval accuracy drops from 79.57 to 79.35 and CoQA F1 actually rises from 72.60 to 75.18. It argues that the early-stopping mechanism prevents overfitting, the single-layer FFN restriction targets where facts are stored, and the scaled, pruned knowledge delta suppresses noisy updates, so the merged model retains newly acquired facts and general abilities simultaneously.

Load-bearing premise

The claim that the method excels at sequential editing rests on treating a single fine-tuning pass over the entire edit set as equivalent to editing facts one after another; the paper never defines an edit order or measures retention across a sequence, so if batch training is not a faithful proxy, the sequential comparison is untested.

Editorial extensions

If this is right

  • If the central claim is right, any dense transformer can be updated with a single extra fine-tuning-and-merge pass, since the recipe makes no architectural changes.
  • Batch edits scale to thousands of facts in one run: the paper reports edit success near 99% on ZsRE with C-Eval within about a point of the base model.
  • The merging hyperparameters give a practical dial: scaling near 0.8 and pruning sparsity near 0.2 balance edit fidelity against general capability, per the paper's parameter analysis.
  • The comparison suggests that locate-then-edit methods (ROME, MEMIT) and LoRA lose most edit success when many facts are injected at once, while the proposed recipe does not.

Reading between the lines

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

  • The 'sequential editing' claim is not actually tested as sequential in the paper: Algorithm 1 iterates over the full dataset, and no edit order, incremental retention measure, or per-fact forgetting check is reported, so a true sequential protocol could change the conclusion.
  • The method is task arithmetic with pruning and rescaling applied to knowledge editing, which opens testable refinements from the merging literature, such as pruning by sign consensus or interpolating multiple edited checkpoints.
  • The paper's own limitation section reports weaker transfer of edited knowledge to rephrased or multi-hop queries, so the recipe is best suited to direct factual updates; a natural extension is paraphrase augmentation or continued fine-tuning on rephrased edits to lift portability.
  • A practical extension of the same merge-after-fine-tune recipe would be continual learning: as new edits arrive, merge against the previously edited checkpoint instead of the base, and measure whether general ability drifts over many rounds.
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 proposes a two-stage framework for knowledge editing in LLMs: first, robust supervised fine-tuning (R-SFT) that iterates over samples with early stopping and updates only the FFN of a single transformer layer; second, model merging via a scaled and pruned task vector (Eq. 6) added to the base model. Experiments on ZsRE and several KnowEdit datasets with Qwen2.5-7B-Instruct compare the method against ROME, MEMIT, LoRA, and plain SFT, reporting edit-success, generalization, portability, locality, fluency, and general-capability benchmarks (C-Eval, CoQA, DROP, SQuAD 2.0, LogiQA). The paper claims that the framework significantly outperforms existing methods in sequential editing while better preserving general model capabilities.

Significance. If the results held, the proposed recipe would be a simple, architecture-agnostic baseline for injecting a large set of factual edits while retaining general capabilities, with practical value. The authors ship code and build on standard toolkits (LLaMA Factory, mergeKit, EasyEdit), which aids reproducibility. However, the central sequential-editing claim is not actually tested by the paper's protocol, and there are internal inconsistencies in the reported numbers. As a result, the significance of the contribution as stated is not established; the method may still serve as a useful batch-editing baseline, but the manuscript's current claims outrun its evidence.

major comments (4)
  1. [§3.2 and Algorithm 1] This is the load-bearing issue.
  2. [Table 1 vs Table 4] The reader's note flagged this, and it is indeed load-bearing for the general-capability preservation claim.
  3. [§3.4 and Figures 2-3] This issue affects the fairness of the comparison with baselines.
  4. [§4.2 and Table 1] This is not a fatal flaw by itself, but it compounds the sequential-editing issue.
minor comments (5)
  1. [Section 2.2, Eq. (6)]
  2. [Section 4.1 and References]
  3. [Table 3(a)]
  4. [Figure 1]
  5. [Section 3.1.3 and Appendix A.2]

Circularity Check

1 steps flagged · score 4.0 of 10

Reported merging performance is partly a fitted hyperparameter selection; no equation-level circularity, but the headline comparison is not an independent prediction.

  1. fitted input called prediction [Section 3.4 'Parameter Analysis (RQ3)', Figures 2-3, and Table 1]
    "Finally, five epochs with six steps per sample provide an optimal compromise. ... Figure 2 and Figure 3 demonstrate that scaling has a more immediate and pronounced impact on model performance, with an optimal setting typically around 0.8 to balance knowledge updates and generalization. In contrast, pruning exhibits a more subtle influence, and a sparsity ratio of 0.2 is generally preferred to minimize interference while preserving core capabilities."

    The 'Merged' results in Table 1 are generated using the settings selected in Section 3.4 (5 epochs, 6 steps, alpha around 0.8, sparsity 0.2). The selection curves in Figures 2-3 and Tables 3a-3b use the same editing metrics (Edit Succ., Generalization, Portability, Locality, Fluency) and the same general-capability benchmark (C-Eval) that are later reported as evidence that the method 'significantly outperforms existing methods.' No held-out validation split is described, so the hyperparameters are effectively fitted to the evaluation metrics and the reported performance is the selected optimum of a sweep over those very metrics. The headline result is therefore partly a fitted input presented as an independent prediction rather than a derivation from first principles.

full rationale

The paper contains no equation-level circularity: Eq. (6) is not derived from the metric definitions, and the claimed R-SFT + merging pipeline is not defined in terms of the evaluation quantities. There is no load-bearing self-citation chain or imported uniqueness theorem. The only substantial circularity concern is the hyperparameter-selection procedure: alpha, sparsity, early-stop threshold, and number of epochs/steps are chosen by inspecting the same benchmark metrics that are subsequently reported as the method's performance, with no explicit validation split. This makes the 'optimal' configuration a fit to the reported benchmarks, inflating the apparent advantage. The separate issue that Algorithm 1 is a batch procedure over the whole dataset rather than a defined sequential-editing loop is a correctness/validity concern about the claim, not a circularity of the derivation; it does not by itself make the method circular. Overall, the central claim is partially fitted but retains independent empirical content, so a moderate circularity score of 4 is appropriate.

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

The paper introduces no new theoretical entities. Its reliance is on fitted hyperparameters (alpha, sparsity, tau, learning rate, layer choice, epochs/steps) and on domain assumptions about layer-localized knowledge and pruning behavior. The absence of a real sequential protocol is the most consequential implicit assumption.

free parameters (6)
  • alpha (merging scaling factor) = 0.8 (optimal)
    Controls the trade-off between edited and general knowledge; chosen based on Figure 2 using the reported test benchmarks.
  • pruning sparsity p = 0.2
    Fraction of knowledge delta parameters pruned; chosen based on Figure 3 using the reported test benchmarks.
  • tau (early-stop loss threshold) = 0.1
    Stops per-sample optimization when loss is below tau; selected to balance edit success and generalization in Table 3a.
  • learning rate = 5e-4
    Selected to maximize edit success in Table 9; lower rates improve locality but reduce success.
  • edited layer index = 5 (claimed optimal; appendix shows layer 6 best)
    The method fixes the edited FFN layer to 5, citing ROME, but Appendix Table 6 shows layers 6 and 7 outperform layer 5.
  • epochs and steps per sample = varies (e.g., 10 epochs x 3 steps for R-SFT; 3 epochs x 10 steps for merged)
    Table 3b shows the reported R-SFT and Merged numbers correspond to different epoch/step settings, which is not explained in the paper.
assumptions (4)
  • domain assumption Factual knowledge in transformer LLMs is localized in the feed-forward networks of a single layer, so editing only that layer suffices.
    Invoked in Section 2.1, citing Meng et al. (2022); this underpins the R-SFT design. The appendix data only partially support it since layer choice changes results.
  • domain assumption Magnitude pruning of the fine-tuning delta selectively removes noise and interference while retaining the edited knowledge.
    Underlies Eq. 6 and the claim of general capability restoration; no mechanistic proof is provided.
  • ad hoc to paper A fixed early-stop loss threshold can be applied uniformly across all samples without harming difficult examples.
    The threshold tau is a hyperparameter fit to the data; the assumption that one threshold works for all samples is introduced for convenience.
  • domain assumption The evaluation protocol treats batch fine-tuning on all edits as a proxy for sequential editing.
    The paper never defines an edit order or sequential evaluation; the claim of sequential superiority rests on this unvalidated assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Model Merging for Knowledge Editing." pith.science (2026). https://pith.science/paper/UQJNFBM5

@misc{pith2026250612384,
  author       = {Pith},
  title        = {Pith review of: Model Merging for Knowledge Editing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UQJNFBM5}},
  note         = {Machine review of arXiv:2506.12384}
}
read the original abstract

Large Language Models (LLMs) require continuous updates to maintain accurate and current knowledge as the world evolves. While existing knowledge editing approaches offer various solutions for knowledge updating, they often struggle with sequential editing scenarios and harm the general capabilities of the model, thereby significantly hampering their practical applicability. This paper proposes a two-stage framework combining robust supervised fine-tuning (R-SFT) with model merging for knowledge editing. Our method first fine-tunes the LLM to internalize new knowledge fully, then merges the fine-tuned model with the original foundation model to preserve newly acquired knowledge and general capabilities. Experimental results demonstrate that our approach significantly outperforms existing methods in sequential editing while better preserving the original performance of the model, all without requiring any architectural changes. Code is available at: https://github.com/Applied-Machine-Learning-Lab/MM4KE.

Figures

Figures reproduced from arXiv: 2506.12384 by the authors.

Figure 1
Figure 1. The illustration of three radar charts demonstrates the performance distribution across multiple tasks. The [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Metrics across different scaling ratios, illustrat [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Metrics across different pruning sparseness, [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 14 canonical work pages

  1. [1]

    Yihan Chen, Dongkuan Zhang, Xiang Wang, Yifan Yang, and Heng Wang. 2023. Dare: ldirect parameter editing for adaptive mode reconfiguration. arXiv preprint arXiv:2310.09570

    work page Pith review arXiv 2023

  2. [2]

    Dheeru Dua, Yizhong Wang, Pradeep Dasigi, Gabriel Stanovsky, Sameer Singh, and Matt Gardner. 2019. https://doi.org/10.18653/v1/N19-1246 DROP : A reading comprehension benchmark requiring discrete reasoning over paragraphs . In Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language...

    doi:10.18653/v1/n19-1246 2019

  3. [3]

    Zichuan Fu, Xiangyang Li, Chuhan Wu, Yichao Wang, Kuicai Dong, Xiangyu Zhao, Mengchen Zhao, Huifeng Guo, and Ruiming Tang. 2024. https://doi.org/10.1145/3698878 A unified framework for multi-domain ctr prediction via large language models . ACM Trans. Inf. Syst. Just Accepted

    work page doi:10.1145/3698878 2024

  4. [4]

    Charles Goddard, Shamane Siriwardhana, Malikeh Ehghaghi, Luke Meyers, et al. 2024. https://aclanthology.org/2024.emnlp-industry.36 Arcee's mergekit: A toolkit for merging large language models . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: EMNLP 2024 - Industry Track, Miami, Florida, USA, November 12-16, 2024 ...

    work page 2024

  5. [5]

    Akshat Gupta, Sidharth Baskaran, and Gopala Anumanchipalli. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.1210 Rebuilding ROME : Resolving model collapse during sequential model editing . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 21738--21744, Miami, Florida, USA. Association for Computational Linguistics

    doi:10.18653/v1/2024.emnlp-main.1210 2024

  6. [6]

    Itay Gur, Wei-Cheng Kao, Elias Polymenakos, and Sujith Ravi. 2023. Task arithmetic in the tangent space: Improved editing of pre-trained models. arXiv preprint arXiv:2305.17651

    arXiv 2023

  7. [7]

    Thomas Hartvigsen, Swami Sankaranarayanan, Hamid Palangi, Yoon Kim, and Marzyeh Ghassemi. 2023. https://openreview.net/forum?id=Oc1SIKxwdV Aging with GRACE : Lifelong model editing with discrete key-value adaptors . In Thirty-seventh Conference on Neural Information Processing Systems

    work page 2023

  8. [8]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685

    arXiv 2021

Show all 44 references
  1. [9]

    Shuyan Huang, Zitao Liu, Xiangyu Zhao, Weiqi Luo, and Jian Weng. 2023 a . https://doi.org/10.1145/3539618.3592073 Towards robust knowledge tracing models via k-sparse attention . In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Infor...

    2023

  2. [10]

    Yuzhen Huang, Yuzhuo Bai, Zhihao Zhu, Junlei Zhang, Jinghan Zhang, Tangjun Su, Junteng Liu, Chuancheng Lv, Yikai Zhang, Jiayi Lei, Yao Fu, Maosong Sun, and Junxian He. 2023 b . C-eval: A multi-level multi-discipline chinese evaluation suite for foundation models. In Advances i...

    2023

  3. [11]

    Zhiwei Huang, Jiacheng Li, Ninghao Ding, Ramesh Nallapati, and Dan Roth. 2021. Instructedit: Learning to edit language models with natural language instructions. arXiv preprint arXiv:2212.10560

    2021 arXiv

  4. [12]

    Zihao Jiang, Xiao Liu, Yibing Zhang, Hao Chen, and Stan Z Li. 2023. Ties: Temporal interference-free editing strategy for continual learning. arXiv preprint arXiv:2310.18356

    2023 arXiv

  5. [13]

    Wei-Cheng Kao, Itay Gur, Elias Polymenakos, Kushal Bansal, and Sujith Ravi. 2023. Slerp: Spherical linear interpolation between neural networks. arXiv preprint arXiv:2305.17493

    2023 arXiv

  6. [14]

    Omer Levy, Minjoon Seo, Eunsol Choi, and Luke Zettlemoyer. 2017. https://doi.org/10.18653/v1/K17-1034 Zero-shot relation extraction via reading comprehension . In Proceedings of the 21st Conference on Computational Natural Language Learning ( C o NLL 2017) , pages 333--342, Va...

    2017 doi

  7. [15]

    Xiaopeng Li, Lixin Su, Pengyue Jia, Xiangyu Zhao, Suqi Cheng, Junfeng Wang, and Dawei Yin. 2023 a . https://arxiv.org/abs/2312.15450 Agent4ranking: Semantic robust ranking via personalized query rewriting using multi-agent llm . Preprint, arXiv:2312.15450

    2023 arXiv

  8. [16]

    Xueyi Li, Youheng Bai, Teng Guo, Zitao Liu, Yaying Huang, Xiangyu Zhao, Feng Xia, Weiqi Luo, and Jian Weng. 2024. https://doi.org/10.24963/ijcai.2024/654 Enhancing length generalization for attention based knowledge tracing models with linear biases . In Proceedings of the Thi...

    2024 doi

  9. [17]

    Zhenyu Li, Zhi Chen, Yeyun Wang, Yue Feng, Jian Li, Dongsheng Zhao, and Ji-Rong Wen. 2023 b . Melo: Memory-efficient llm optimization. arXiv preprint arXiv:2312.02428

    2023 arXiv

  10. [18]

    Jian Liu, Leyang Cui, Hanmeng Liu, Dandan Huang, Yile Wang, and Yue Zhang. 2020. https://doi.org/10.24963/ijcai.2020/501 Logiqa: A challenge dataset for machine reading comprehension with logical reasoning . In Proceedings of the Twenty-Ninth International Joint Conference on ...

    2020 doi

  11. [19]

    Qidong Liu, Xian Wu, Wanyu Wang, Yejing Wang, Yuanshao Zhu, Xiangyu Zhao, Feng Tian, and Yefeng Zheng. 2025. https://doi.org/10.1609/aaai.v39i11.33327 Llmemb: Large language model can be a good embedding generator for sequential recommendation . Proceedings of the AAAI Confere...

    2025 doi

  12. [20]

    Qidong Liu, Xian Wu, Yejing Wang, Zijian Zhang, Feng Tian, Yefeng Zheng, and Xiangyu Zhao. 2024. http://papers.nips.cc/paper\_files/paper/2024/hash/2f0728449cb3150189d765fc87afc913-Abstract-Conference.html LLM-ESR: large language models enhancement for long-tailed sequential r...

    2024

  13. [21]

    Zitao Liu, Qiongqiong Liu, Teng Guo, Jiahao Chen, Shuyan Huang, Xiangyu Zhao, Jiliang Tang, Weiqi Luo, and Jian Weng. 2023. https://proceedings.neurips.cc/paper_files/paper/2023/file/67fc628f17c2ad53621fb961c6bafcaf-Paper-Datasets_and_Benchmarks.pdf Xes3g5m: A knowledge tracin...

    2023

  14. [22]

    Jinliang Lu, Ziliang Pang, Min Xiao, Yaochen Zhu, Rui Xia, and Jiajun Zhang. 2024. https://arxiv.org/abs/2407.06089 Merge, ensemble, and cooperate! a survey on collaborative strategies in the era of large language models . Preprint, arXiv:2407.06089

    2024 arXiv

  15. [23]

    Yun Luo, Zhen Yang, Fandong Meng, Yafu Li, Jie Zhou, and Yue Zhang. 2023. https://doi.org/10.48550/ARXIV.2308.08747 An empirical study of catastrophic forgetting in large language models during continual fine-tuning . CoRR, abs/2308.08747

    2023 doi arXiv

  16. [24]

    Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. 2022. http://papers.nips.cc/paper\_files/paper/2022/hash/6f1d43d5a82a37e89b0665b33bf3a182-Abstract-Conference.html Locating and editing factual associations in GPT . In Advances in Neural Information Processing System...

    2022

  17. [26]

    Eric Mitchell, Kee Siew Lee, Hao Chen, Kevin Meng, David Bau, and Yonatan Belinkov. 2022 a . Memory editing via model editing: Memory editing in large language models. arXiv preprint arXiv:2210.07229

    2022 arXiv

  18. [27]

    Eric Mitchell, Charles Lin, Antoine Bosselut, Chelsea Finn, and Christopher D Manning. 2021. Fast model editing at scale. In International Conference on Learning Representations

    2021

  19. [28]

    Eric Mitchell, Charles Lin, Antoine Bosselut, Christopher D Manning, and Chelsea Finn. 2022 b . Memory-based model editing at scale. In International Conference on Machine Learning, pages 15817--15831. PMLR

    2022

  20. [29]

    OpenAI, Josh Achiam, Steven Adler, et al. 2024. https://arxiv.org/abs/2303.08774 Gpt-4 technical report . Preprint, arXiv:2303.08774

    2024 arXiv

  21. [30]

    Pranav Rajpurkar, Robin Jia, and Percy Liang. 2018. https://doi.org/10.18653/v1/P18-2124 Know what you don`t know: Unanswerable questions for SQ u AD . In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 78...

    2018 doi

  22. [31]

    Siva Reddy, Danqi Chen, and Christopher D. Manning. 2019. https://doi.org/10.1162/tacl_a_00266 C o QA : A conversational question answering challenge . Transactions of the Association for Computational Linguistics, 7:249--266

    2019 doi

  23. [32]

    Song Wang, Yaochen Zhu, Haochen Liu, Zaiyi Zheng, Chen Chen, and Jundong Li. 2024 a . https://doi.org/10.1145/3698590 Knowledge editing for large language models: A survey . ACM Comput. Surv., 57(3)

    2024 doi

  24. [33]

    Zihao Wang, Yihua Chen, Hao Xie, Xiang Li, Ningyu Zhang, and Huajun Chen. 2024 b . Wise: Memory-efficient model editing with task-aware compression. arXiv preprint arXiv:2401.12174

    2024 arXiv

  25. [34]

    Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Kornblith, and Ludwig Schmidt

    Mitchell Wortsman, Gabriel Ilharco, Samir Yitzhak Gadre, Rebecca Roelofs, Raphael Gontijo Lopes, Ari S. Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Kornblith, and Ludwig Schmidt. 2022. https://proceedings.mlr.press/v162/wortsman22a.html Model soups: averaging we...

    2022

  26. [35]

    Derong Xu, Ziheng Zhang, Zhenxi Lin, Xian Wu, Zhihong Zhu, Tong Xu, Xiangyu Zhao, Yefeng Zheng, and Enhong Chen. 2024 a . https://aclanthology.org/2024.lrec-main.1044/ Multi-perspective improvement of knowledge graph completion with large language models . In Proceedings of th...

    2024

  27. [36]

    Derong Xu, Ziheng Zhang, Zhihong Zhu, Zhenxi Lin, Qidong Liu, Xian Wu, Tong Xu, Wanyu Wang, Yuyang Ye, Xiangyu Zhao, Enhong Chen, and Yefeng Zheng. 2024 b . https://doi.org/10.1145/3627673.3679673 Editing factual knowledge and explanatory ability of medical large language mode...

    2024

  28. [37]

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jianxin Yang, Jin Xu, Jingren Zhou, Jinze...

    2024 arXiv

  29. [38]

    Ningyu Zhang, Bozhong Tian, Siyuan Cheng, Xiaozhuan Liang, Yi Hu, Kouying Xue, Yanjie Gou, Xi Chen, and Huajun Chen. 2024 a . https://doi.org/10.24963/ijcai.2024/733 Instructedit: instruction-based knowledge editing for large language models . In Proceedings of the Thirty-Thir...

    2024 doi

  30. [40]

    Ningyu Zhang, Yunzhi Yao, Bozhong Tian, et al. 2024 c . https://arxiv.org/abs/2401.01286 A comprehensive study of knowledge editing for large language models . Preprint, arXiv:2401.01286

    2024 arXiv

  31. [41]

    Wayne Xin Zhao, Kun Zhou, Junyi Li, et al. 2024. https://arxiv.org/abs/2303.18223 A survey of large language models . Preprint, arXiv:2303.18223

    2024 arXiv

  32. [42]

    Xiangyu Zhao, Long Xia, Liang Zhang, Zhuoye Ding, Dawei Yin, and Jiliang Tang. 2018 a . https://doi.org/10.1145/3240323.3240374 Deep reinforcement learning for page-wise recommendations . In Proceedings of the 12th ACM Conference on Recommender Systems, RecSys '18, page 95–103...

    2018

  33. [43]

    Xiangyu Zhao, Liang Zhang, Zhuoye Ding, Long Xia, Jiliang Tang, and Dawei Yin. 2018 b . https://doi.org/10.1145/3219819.3219886 Recommendations with negative feedback via pairwise deep reinforcement learning . In Proceedings of the 24th ACM SIGKDD International Conference on K...

    2018

  34. [44]

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, and Yongqiang Ma. 2024. https://doi.org/10.48550/ARXIV.2403.13372 Llamafactory: Unified efficient fine-tuning of 100+ language models . CoRR, abs/2403.13372

    2024 doi arXiv

  35. [45]

    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...

  36. [46]

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