Pith. sign in

REVIEW 3 major objections 4 minor 24 references

LP-SFT shows that supervised fine-tuning can reduce catastrophic forgetting and preserve generation diversity by adding a locally normalized KL term over the frozen base model's top-10 non-target token alternatives, alongside standard cross

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 08:32 UTC pith:4THWGCRD

load-bearing objection A cheap, effective SFT regularizer with consistent gains, but the entropy-based motivation is softer than advertised and the experiments need error bars before I'd trust the aggregate numbers. the 3 major comments →

arxiv 2607.04733 v2 pith:4THWGCRD submitted 2026-07-06 cs.CL cs.LG

LP-SFT: Local-Preserving Supervised Fine-Tuning via Multimodal Entropy Structure

Yueyang Wang , Baolong Bi , Shuo Lu , Jingyuan Zhang , Jiajun Shi This is my paper
classification cs.CL cs.LG
keywords supervised fine-tuningcatastrophic forgettingmultimodal entropy structurelocal preservationKL divergencetop-k supportpass@k diversitylanguage models
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Standard supervised fine-tuning (SFT) improves a model on a target task but often degrades existing abilities and shrinks output diversity, because cross-entropy trains only the observed label token and leaves unconstrained how probability mass is redistributed over other plausible tokens. This paper shows that pretrained models encode a regular multimodal entropy structure: next-token distributions contain near-uniform plateaus at k = 1, 2, ..., 10 plausible alternatives, reflecting logical branching at low order and structural ambiguity (digits, dates, IDs) at high order. The paper proposes LP-SFT, which adds to cross-entropy a locally normalized KL divergence over the frozen base model's top-10 non-target candidates, preserving relative preferences among those alternatives without constraining their absolute probability mass. Across mixed-domain and single-domain fine-tuning experiments, LP-SFT outperforms vanilla SFT and recent SFT-enhancement baselines, achieving the best balance between pass@1 accuracy and pass@k diversity while retaining general knowledge. If correct, local preservation is a cheap and effective way to mitigate catastrophic forgetting and maintain sampling-accessible diversity during fine-tuning.

Core claim

The paper's central claim is that the harmful side effects of supervised fine-tuning arise from overwriting the pretrained model's local preference structure, and that this structure can be preserved at almost no extra computational cost. The authors first demonstrate that base models exhibit next-token entropy peaks at H ≈ ln k for k = 1 through 10 and beyond, meaning token positions fall into discrete uncertainty regimes: some are nearly deterministic, others admit a small set of roughly equally plausible continuations. Standard cross-entropy, rewarding only the observed target token, suppresses these alternatives and distorts the pretrained structure, which the authors identify as a mecha

What carries the argument

The central mechanism is the multimodal entropy structure of next-token distributions, quantified by the effective support ratio R = N2/N1, where N1 = exp(H1) is the Shannon effective support and N2 = exp(H2) the Rényi-2 effective support over a normalized top-30 window. Peaks in the (N1, R) plane near integer N1 with R close to 1 identify plateau-k states — positions where uncertainty is spread nearly uniformly across k plausible alternatives. LP-SFT uses this structure to build its preservation set: at each position it takes the frozen base model's top-10 tokens (Kmax = 10, chosen to cover low-order plateaus and the prominent k ≈ 10 regime), removes the supervised target token, and applies

Load-bearing premise

The load-bearing premise is that a fixed top-10 truncation of the frozen base distribution captures the alternatives whose preservation protects pre-trained capabilities; if the effective local support is much larger or concentrated above k = 10, the preservation term omits what matters and the method's benefit could shrink or even reverse.

What would settle it

Measure the effective support size N1 of a base model's next-token distributions at positions where fine-tuning causes the largest accuracy drops; if most of those positions have N1 well above 10 (say, >20), then LP-SFT with Kmax = 10 should fail to preserve them, and a version with larger or adaptive K should outperform it. A direct experiment is to fine-tune with Kmax = 10 versus Kmax = 30 or 100 and compare pass@1, pass@k, and MMLU; a substantial gain from larger K would indicate that the fixed 10-token truncation is the bottleneck.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If LP-SFT is right, then standard SFT's catastrophic forgetting is largely a failure to preserve local preference structure, and a lightweight top-10 constraint can recover much of the benefit of full-vocabulary anchoring at a fraction of the computational cost.
  • General knowledge (as measured by MMLU) degrades far less under LP-SFT than under vanilla cross-entropy or other SFT variants, narrowing the average gap to the frozen base model from double digits to about one point in the mixed-domain setting.
  • Pass@k diversity is maintained or improved while pass@1 accuracy also improves, suggesting that preserving non-label alternatives does not come at the expense of single-sample correctness.
  • LP-SFT is relatively insensitive to the preservation weight μ, unlike full-vocabulary KL anchoring which is reported to be sensitive to its KL weight; this makes the method easier to deploy without hyperparameter tuning.
  • The method's benefit depends on the quality of the base distribution: for weaker backbones on math tasks, vanilla cross-entropy can remain stronger on some metrics, implying that local preservation is most helpful when the pretrained distribution is a reliable structural reference.

Where Pith is reading between the lines

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

  • Because LP-SFT's preservation relies on the frozen base distribution being a good structural reference, the analysis suggests that fine-tuning data containing many structurally ambiguous positions (digits, dates, IDs) will benefit most; these are exactly the positions where cross-entropy injects arbitrary supervision.
  • The plateau-k framework could be used as a diagnostic: measuring N1 and R before fine-tuning might allow one to predict which domains will suffer the most forgetting under vanilla SFT and to choose Kmax accordingly, rather than fixing it at 10.
  • A testable extension is to let Kmax adapt per position to the base distribution's observed effective support; this should improve retention on models or domains where the support exceeds 10, as hinted by the paper's own weaker-backbone math results.
  • The local KL term suggests a broader principle: when adapting a model, preserve the relative ordering of plausible alternatives, not just the top-1. This principle could transfer to preference optimization or RLHF, where reference-policy structure is often regularized globally rather than locally.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes LP-SFT, a supervised fine-tuning objective that adds a local KL-regularization term to cross-entropy. The term is computed over a frozen base model's top-K candidate set after removing the supervised target token, with locally normalized probabilities, intended to preserve the base model's relative preferences among non-label alternatives. The motivation is an empirical analysis of next-token distributions reporting a multimodal 'plateau-k' entropy structure, including higher-order plateaus around k=10, 12, and 15. Experiments fine-tune Qwen3-4B, Qwen3-14B, and Llama-3.1-8B on UltraFeedback, Magicoder-OSS-Instruct-75K, and NuminaMath-CoT, and evaluate on MATH-500, AIME 2021-26, MBPP+, HumanEval+, and MMLU. The paper reports that LP-SFT improves aggregate pass@1/pass@k and MMLU retention over vanilla CE, DFT, EAFT, GEM, and ASFT, with ablations supporting the target-token-removal and local-normalization design choices.

Significance. If the empirical claims are robust, LP-SFT is a practical, low-overhead regularization scheme for supervised fine-tuning that mitigates catastrophic forgetting and preserves sampling diversity without full-vocabulary anchoring. The paper's strengths include evaluation across multiple model families and scales, multiple training distributions, and several recent baselines; a careful ablation of the two design choices; a computationally efficient offline base-model cache; and released code. The central weakness is that all comparisons are single-seed point estimates with no variance reporting, and the fixed truncation at Kmax=10 is not justified by the paper's own entropy analysis, which documents plateau states beyond that cutoff. The motivation-to-method link is therefore not fully established.

major comments (3)
  1. [§5.1, §5.2, Appendix B] The method hard-truncates the preservation set at Kmax=10, but the paper's own analysis documents plateau-k states at k=12 (Table 12: N1=11.998, R=0.9981) and k=15 (Table 13: N1=14.966, R=0.9969), and Figure 1 shows ridges beyond N1=10. The text in §5.2 explicitly states that the local set 'may not cover the entire plateau region, especially when the effective support of the base distribution exceeds Kmax.' Since high-order plateaus are argued to represent structural ambiguity (random IDs, URLs, digits, timestamps), domains rich in such tokens are exactly where LP-SFT's protection is incomplete. The claim that LP-SFT preserves the multimodal entropy structure identified in Section 4 is therefore only partially realized. Please either (a) provide evidence that k>10 plateaus are negligible for retention/diversity, or (b) include a sensitivity analysis over Kmax (e.g., 10, 15, 20) showing t
  2. [§6.1, Appendix C.1, Tables 1, 2, 15, 16] All experiments use a single fixed seed (1234) and report point estimates without error bars, confidence intervals, or multiple runs. Several headline differences are numerically small; for example, Table 2 (Llama-3.1-8B, UltraFeedback) shows LP-SFT Avg 33.91 vs. CE 33.20, and Table 16 (Qwen3-4B, NuminaMath) shows Avg 61.25 vs. CE 60.35. The central claim that LP-SFT 'improves overall performance' is not statistically supported. Please run at least 3 seeds for the main comparisons (or provide bootstrap intervals over evaluation samples) and report mean±std, so that the reader can judge the significance of the reported gains.
  3. [§6.3, Tables 2 and 16] The abstract and conclusion state that LP-SFT improves overall performance across mixed-domain and single-domain experiments, but the results for Llama-3.1-8B show that CE is stronger on several math metrics (Table 2: MATH-500 pass@1 10.76 vs. 9.41; AIME pass@1 0.23 vs. 0.12), and the overall advantage is small. The discussion acknowledges this in prose, but the general claim should be qualified or the paper should characterize the conditions under which local preservation helps versus hurts. One concrete step is to analyze the base model's entropy structure for Llama-3.1-8B and relate it to the observed performance pattern; currently the reader cannot tell whether the method's benefit depends on the quality of the base distribution's plateau structure.
minor comments (4)
  1. [Abstract/Title] The title has typographical spacing errors: 'LOCAL-PRESERVINGSUPERVISEDFINE-TUNING' and 'VIAMULTIMODALENTROPYSTRUCTURE' should be separated into words.
  2. [§5.1] The notation 'K=K max=10' is confusing because Section 3 uses K=30 for the entropy-analysis truncation. Please use distinct symbols, e.g., K_ent for analysis and K_max for the preservation set, and state the default values explicitly.
  3. [Appendix E.3, Table 18] Training-time ratios exclude the offline base-model precomputation pass. While the text explains that this pass is amortized, the first-run end-to-end time should also be reported, since Stage 1 is a full forward pass over the training set and may be nontrivial for large models or large datasets.
  4. [§4.2, Eq. (12)] The plateau-k definition uses fixed thresholds N1 in [k-0.3, k+0.3] and R>0.95. This is a reasonable heuristic, but the paper should state how sensitive the observed conclusions in Figures 1-2 are to these thresholds, or at least acknowledge that the thresholds are chosen post hoc.

Circularity Check

0 steps flagged

No significant circularity: LP-SFT's headline benchmark gains are empirical, not entailed by its preservation objective; only a non-load-bearing self-citation and an internal design link keep it from a clean 0.

full rationale

The derivation chain is observation (Section 4) -> objective (Section 5) -> measured benchmarks (Section 6). The local-preservation objective in Eqs. (13)-(17) is literally a KL match to the frozen base model's conditional distribution on a truncated, target-removed set; saying it 'preserves local preferences' is a description of the loss, not a derived prediction. The paper's substantive claims (Tables 1-2 pass@1/pass@k, MMLU retention) come from held-out evaluations against CE, DFT, EAFT, GEM, and ASFT. No benchmark number is used to select Kmax=10 or µ=1, so there is no fitted-input-called-prediction. The multimodal-entropy motivation is reproduced in Section 4 on Qwen3-4B, Llama-3.1-8B, and Qwen3-30B-A3B rather than imported solely from ref [19], which is authored by overlapping authors but is not load-bearing. The paper's own limitation that plateau states exist beyond Kmax (Appendix B, k=12 and k=15; §5.2) weakens the coverage claim and is a robustness/correctness issue, not a circularity. The ablation showing a full-vocabulary variant performs comparably (Table 17) also indicates the local design is an empirical choice, not a definitional identity.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claim rests on an empirical method rather than a derivation, so there are no invented physical entities. The free parameters are design choices; the key axioms are interpretative and methodological assumptions about entropy structure and preservation.

free parameters (4)
  • Kmax (preservation support size) = 10
    Chosen from the entropy analysis because of the prominent plateau around k=10; central design hyperparameter, not swept in experiments.
  • mu (preservation weight) = 1
    Default preservation strength; ablation shows insensitivity over [0.1,1.0], but it is still a hand-set weight.
  • K_trunc (entropy analysis truncation) = 30
    Fixed top-K window used to compute N1 and N2; hand-chosen and used for the lower-envelope derivation.
  • Plateau-k thresholds = N1 in [k-0.3, k+0.3], R > 0.95
    Ad hoc definitions of entropy-compressed states; used to justify Kmax=10 and the low/high-order separation.
axioms (5)
  • standard math Jensen's inequality and Lagrangian stationarity justify R<=1 and the two-level extremal family for the lower envelope.
    Used in Section 4.1 and Appendix A to characterize the finite-top-K lower envelope.
  • domain assumption Entropy peaks at ln k correspond to genuinely plausible alternatives rather than tokenization artifacts.
    Section 4.2 interprets plateau-k states as logical branching or structural ambiguity; this is the paper's core interpretative premise.
  • domain assumption Preserving relative preferences among non-label top-K alternatives mitigates forgetting without hurting target-token learning.
    Core hypothesis of the LP-SFT objective in Section 5; supported empirically but not proven mechanistically.
  • domain assumption The frozen base model's local distribution is a reliable structural reference across domains and scales.
    Required for the method; the paper itself notes weaker gains for Llama-3.1-8B, partially supporting the assumption's limits.
  • domain assumption Pass@k measures sampling-accessible diversity.
    Used throughout Section 6 as the diversity-oriented metric alongside pass@1.

pith-pipeline@v1.3.0-alltime-deepseek · 18481 in / 13206 out tokens · 140049 ms · 2026-08-02T08:32:53.885955+00:00 · methodology

0 comments
read the original abstract

Supervised fine-tuning (SFT) is the standard approach for adapting pretrained language models to downstream domains, yet it often improves target-domain behavior at the cost of degrading pre-existing capabilities. Standard cross-entropy fine-tuning promotes only the observed label token and leaves unconstrained how probability mass is redistributed over other plausible alternatives, potentially distorting the rich local preference structure learned during pretraining. We first analyze next-token predictions using Shannon and Renyi entropies, revealing that pretrained models exhibit a regular multimodal entropy structure. These entropy peaks correspond to varying numbers of plausible alternatives, indicating that the base model intrinsically encodes rich distributional knowledge beyond the single supervised token. Motivated by this observation, we propose LP-SFT, a Local-Preserving Supervised Fine-Tuning objective designed to explicitly protect this inherent entropy structure. At each step, LP-SFT constructs a local top-K support of alternative tokens from the frozen base distribution. Crucially, it removes the supervised target token from this set to avoid conflicting with the cross-entropy objective, and applies a locally normalized KL divergence to maintain the base model's relative preference structure among the remaining non-label alternatives. Across mixed-domain and single-domain fine-tuning experiments, LP-SFT improves overall performance over vanilla SFT and recent SFT-enhancement baselines, achieving the best balance between pass@1 accuracy and pass@k performance. These results suggest that local preservation helps mitigate capability degradation without collapsing sampling-accessible diversity.

Figures

Figures reproduced from arXiv: 2607.04733 by Baolong Bi, Jiajun Shi, Jingyuan Zhang, Shuo Lu, Yueyang Wang.

Figure 1
Figure 1. Figure 1: Joint distributions of N1 and R = N2/N1. High-R ridges at multiple integer supports reveal higher￾order entropy peaks across datasets and model families. Since N1(pt) = exp(H1(pt)), an entropy peak at H1(pt) ≈ ln m is equivalently observed as a concentration near N1(pt) ≈ m. In the ideal plateau case where the normalized next-token distribution is uniform over m tokens, we have H1(pt) = H2(pt) = ln m, N1(p… view at source ↗
Figure 2
Figure 2. Figure 2: Occurrence of plateau-k states, defined by N1 ∈ [k − 0.3, k + 0.3] and R > 0.95. Instruction tuning increases low-order plateaus (k = 1–5) while reducing or preserving higher-order plateaus (k > 5). 4.2 Plateau-k States: From Low-Order Branching to High-Order Ambiguity We define entropy-compressed states of order k as plateau-k states, captured by N1 ∈ [k − 0.3, k + 0.3] and R > 0.95. (12) [PITH_FULL_IMAG… view at source ↗
Figure 3
Figure 3. Figure 3: Qwen3-14B fine-tuning results across three training datasets. the best pass@k performance across all displayed sampling-based evaluations, suggesting that it improves single-sample correctness without collapsing sampling-accessible diversity. Compared with the base model, LP-SFT also shows the smallest overall degradation among fine-tuned methods, reducing the Avg. gap from 7.50 points under CE to 2.72 poi… view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity to the preservation weight µ. Results are obtained with Qwen3-4B fine-tuned on UltraFeedback and evaluated on MATH-500, MBPP+, and MMLU. Performance varies only slightly across µ ∈ [0.1, 1.0], suggesting that LP-SFT is relatively insensitive to the choice of µ [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗

Review history (2 revisions) →

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

24 extracted references · 10 linked inside Pith

  1. [1]

    On-policy distillation of language models: Learning from self-generated mistakes

    Rishabh Agarwal, Nino Vieillard, Yongchao Zhou, Piotr Stanczyk, Sabela Ramos Garea, Matthieu Geist, and Olivier Bachem. On-policy distillation of language models: Learning from self-generated mistakes. In International Conference on Learning Representations, volume 2024, pages 21246–21263, 2024

    2024

  2. [2]

    Evaluating large language models trained on code.arXiv preprint arXiv:2107.03374, 2021

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde De Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. Evaluating large language models trained on code.arXiv preprint arXiv:2107.03374, 2021

    Pith/arXiv arXiv 2021

  3. [3]

    Ultrafeedback: Boosting language models with scaled ai feedback.arXiv preprint arXiv:2310.01377, 2023

    Ganqu Cui, Lifan Yuan, Ning Ding, Guanming Yao, Bingxiang He, Wei Zhu, Yuan Ni, Guotong Xie, Ruobing Xie, Yankai Lin, et al. Ultrafeedback: Boosting language models with scaled ai feedback.arXiv preprint arXiv:2310.01377, 2023. 10

    Pith/arXiv arXiv 2023

  4. [4]

    Entropy-adaptive fine-tuning: Resolving confident conflicts to mitigate forgetting

    Muxi Diao, Lele Yang, Wuxuan Gong, Yutong Zhang, Zhonghao Yan, Yufei Han, Kongming Liang, Weiran Xu, and Zhanyu Ma. Entropy-adaptive fine-tuning: Resolving confident conflicts to mitigate forgetting. arXiv preprint arXiv:2601.02151, 2026

    arXiv 2026

  5. [5]

    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

    Pith/arXiv arXiv 2024

  6. [6]

    Minillm: Knowledge distillation of large language models

    Yuxian Gu, Li Dong, Furu Wei, and Minlie Huang. Minillm: Knowledge distillation of large language models. InInternational Conference on Learning Representations, volume 2024, pages 32694–32717, 2024

    2024

  7. [7]

    Measuring massive multitask language understanding.arXiv preprint arXiv:2009.03300, 2020

    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

    Pith/arXiv arXiv 2009

  8. [8]

    Measuring mathematical problem solving with the math dataset.arXiv preprint arXiv:2103.03874, 2021

    Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathematical problem solving with the math dataset.arXiv preprint arXiv:2103.03874, 2021

    Pith/arXiv arXiv 2021

  9. [9]

    Diversity and evenness: a unifying notation and its consequences.Ecology, 54(2):427–432, 1973

    Mark O Hill. Diversity and evenness: a unifying notation and its consequences.Ecology, 54(2):427–432, 1973

    1973

  10. [10]

    Distilling the knowledge in a neural network.arXiv preprint arXiv:1503.02531, 2015

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network.arXiv preprint arXiv:1503.02531, 2015

    Pith/arXiv arXiv 2015

  11. [11]

    Entropy and diversity.Oikos, 113(2):363–375, 2006

    Lou Jost. Entropy and diversity.Oikos, 113(2):363–375, 2006

    2006

  12. [12]

    Numinamath: The largest public dataset in ai4maths with 860k pairs of competition math problems and solutions.Hugging Face repository, 13(9):9, 2024

    Jia Li, Edward Beeching, Lewis Tunstall, Ben Lipkin, Roman Soletskyi, Shengyi Huang, Kashif Rasul, Longhui Yu, Albert Q Jiang, Ziju Shen, et al. Numinamath: The largest public dataset in ai4maths with 860k pairs of competition math problems and solutions.Hugging Face repository, 13(9):9, 2024

    2024

  13. [13]

    Preserving diversity in supervised fine-tuning of large language models

    Ziniu Li, Congliang Chen, Tian Xu, Zeyu Qin, Jiancong Xiao, Zhi-Quan Luo, and Ruoyu Sun. Preserving diversity in supervised fine-tuning of large language models. InInternational Conference on Learning Representations, volume 2025, pages 66127–66154, 2025

    2025

  14. [14]

    Jiawei Liu, Chunqiu Steven Xia, Yuyao Wang, and Lingming Zhang. Is your code generated by chatgpt really correct? rigorous evaluation of large language models for code generation.Advances in neural information processing systems, 36:21558–21572, 2023

    2023

  15. [15]

    An empirical study of catastrophic forgetting in large language models during continual fine-tuning.IEEE Transactions on Audio, Speech and Language Processing, 2025

    Yun Luo, Zhen Yang, Fandong Meng, Yafu Li, Jie Zhou, and Yue Zhang. An empirical study of catastrophic forgetting in large language models during continual fine-tuning.IEEE Transactions on Audio, Speech and Language Processing, 2025

    2025

  16. [16]

    Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    2022

  17. [17]

    Shenzhi Wang, Le Yu, Chang Gao, Chujie Zheng, Shixuan Liu, Rui Lu, Kai Dang, Xiong-Hui Chen, Jianxin Yang, Zhenru Zhang, et al. Beyond the 80/20 rule: High-entropy minority tokens drive effective reinforcement learning for llm reasoning.Advances in Neural Information Processing Systems, 38:115452– 115486, 2026

    2026

  18. [18]

    Self-instruct: Aligning language models with self-generated instructions

    Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A Smith, Daniel Khashabi, and Hannaneh Hajishirzi. Self-instruct: Aligning language models with self-generated instructions. In Proceedings of the 61st annual meeting of the association for computational linguistics (volume 1: long papers), pages 13484–13508, 2023

    2023

  19. [19]

    He-snr: Uncovering latent logic via entropy for guiding mid-training on swe-bench.arXiv preprint arXiv:2601.20255, 2026

    Yueyang Wang, Jiawei Fu, Baolong Bi, Xili Wang, and Xiaoqing Liu. He-snr: Uncovering latent logic via entropy for guiding mid-training on swe-bench.arXiv preprint arXiv:2601.20255, 2026

    Pith/arXiv arXiv 2026

  20. [20]

    Magicoder: Empowering code generation with oss-instruct.arXiv preprint arXiv:2312.02120, 2023

    Yuxiang Wei, Zhe Wang, Jiawei Liu, Yifeng Ding, and Lingming Zhang. Magicoder: Empowering code generation with oss-instruct.arXiv preprint arXiv:2312.02120, 2023

    Pith/arXiv arXiv 2023

  21. [21]

    Neural text generation with unlikelihood training.arXiv preprint arXiv:1908.04319, 2019

    Sean Welleck, Ilia Kulikov, Stephen Roller, Emily Dinan, Kyunghyun Cho, and Jason Weston. Neural text generation with unlikelihood training.arXiv preprint arXiv:1908.04319, 2019

    Pith/arXiv arXiv 1908

  22. [22]

    On the generalization of sft: A reinforcement learning perspective with reward rectification.arXiv preprint arXiv:2508.05629, 2025

    Yongliang Wu, Yizhou Zhou, Zhou Ziheng, Yingzhe Peng, Xinyu Ye, Xinting Hu, Wenbo Zhu, Lu Qi, Ming-Hsuan Yang, and Xu Yang. On the generalization of sft: A reinforcement learning perspective with reward rectification.arXiv preprint arXiv:2508.05629, 2025

    arXiv 2025

  23. [23]

    Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025. 11

    Pith/arXiv arXiv 2025

  24. [24]

    Risk perceptions and safety compliance of workers employed in agriculture, forestry, and fishing

    He Zhu, Junyou Su, Peng Lai, Ren Ma, Wenjia Zhang, Linyi Yang, and Guanhua Chen. Anchored supervised fine-tuning.arXiv preprint arXiv:2509.23753, 2025. A Derivation of the Finite-top-KLower Envelope We derive the finite-top-K lower-envelope branch of R for a given Shannon effective support size N1. Let ˆp= (ˆp1, . . . ,ˆpK) denote the normalized top- K ne...

    arXiv 2025