Pith. sign in

REVIEW 3 major objections 5 minor 3 cited by

Enhancing the Reasoning Capabilities of Small Language Models via Solution Guidance Fine-Tuning

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read With only 3,000 solution-guidance examples, small language models outperform chain-of-thought fine-tuning with 30,000 examples on five reasoning benchmarks.

desk verdict A practical plan-and-execute distillation recipe with a strong but unproven data-efficiency claim. read the letter →

arxiv 2412.09906 v1 pith:VQEM72DG submitted 2024-12-13 cs.CL

Jing Bi , Yuting Wu , Weiwei Xing , Zhenjie Wei This is my paper · ORCID
classification cs.CL
keywords solutionguidancesmalllanguagemodelsreasoningchain-of-thoughtfine-tuningcollaborativeinferencedataefficiencyGSM8KLISA
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 claims that the reasoning bottleneck for small language models (under 10 billion parameters) is problem decomposition, not computation, and that a model trained to emit concise, calculation-free solution plans can guide any other small model to the correct answer. It introduces Solution Guidance (SG), a data format of two-to-six step solution objectives, and Solution-Guidance Fine-Tuning (SGFT), which trains a small model on 1,000-3,000 SG examples to produce such plans. In collaborative inference, the fine-tuned planner's guidance is passed, together with the original question, to a second, unmodified small model that writes the final answer. Across GSM8K, SVAMP, MultiArith, StrategyQA, and CommonsenseQA, the approach outperforms chain-of-thought fine-tuning with 30,000 examples, and the paper reports that 1,000 SG examples already beat 30,000 CoT examples, cutting data requirements to roughly 3-10%. If correct, it makes stronger reasoning practical on a single consumer-grade GPU without large annotated CoT corpora.

What carries the argument

The load-bearing object is the Solution Guidance (SG) data schema: a question paired with a two-to-six step list of solution objectives and their order, explicitly without calculations or final answers. The protocol that carries the result is two-model collaborative inference: a fine-tuned guidance model (Model_SG) reads the question and emits SG, and an untrained response model receives the original question plus SG and generates the answer. Fine-tuning uses the LISA layer-wise importance sampling method on a consumer GPU, with contextual prompts (zero-shot or few-shot) included in the training data to stop the planner from slipping into calculations. The division of labor is what the paper credits for the accuracy gain, since an SG-fine-tuned planner paired with an untrained answerer outperforms all other combinations in their ablation, including pairing the planner with a CoT-fine-tuned answerer.

What would settle it

Fine-tune a small model on 3,000 GPT-4o-generated Chain-of-Thought examples instead of Solution Guidance, keep the same two-model collaborative inference (the fine-tuned model as planner, an untrained model as answerer), and evaluate on GSM8K: if the CoT-based planner matches or beats the SG-based planner's 43.7 accuracy, the SG format is not the decisive factor; if it falls short, the format's contribution is confirmed.

Watch

Extended reading notes

Core claim

The central discovery is that separating planning from calculation improves small-model reasoning. A small language model fine-tuned on Solution Guidance, which lists step-by-step solution objectives without any arithmetic, can decompose unseen problems into generic plans, and a second, unmodified small model can execute those plans to produce final answers. The paper reports that SGFT with 3,000 SG examples, generated from GSM8K training questions by GPT-4o, beats CoT fine-tuning with 30,000 examples across all five tested benchmarks; for instance, ChatGLM3-6B guided by ChatGLM3-6B_SG reaches 43.7 on GSM8K versus 34.4 for the CoT-fine-tuned model, and the best cross-model pair, Qwen2-7B_SG generating guidance for ChatGLM3-6B, reaches 48.3. The authors argue this works because SG focuses on problem understanding and decomposition at the semantic and logical levels, avoiding the error propagation and noise that come from concatenating calculations in CoT chains.

Load-bearing premise

The paper's central claim rests on the assumption that the solution-guidance data format, rather than the stronger GPT-4o teacher or the two-model inference setup, is what drives the accuracy gains.

Editorial extensions

If this is right

  • With roughly 3,000 SG examples, practitioners can fine-tune a 6-7B model on a single consumer-grade GPU and exceed CoT fine-tuning that uses 30,000 examples, cutting data preparation cost by an order of magnitude.
  • Because the answer model needs no training, any small instruction-tuned model can serve as the responder, so a single SG planner can be reused with different answer models depending on deployment needs.
  • The guidance format transfers across benchmarks: training data drawn only from GSM8K improves out-of-distribution performance on SVAMP, MultiArith, StrategyQA, and CommonsenseQA, suggesting the learned plans capture reusable problem-solving patterns.
  • Separating planning from calculation keeps the planner from generating arithmetic errors, and the paper's ablation indicates that pairing an SG planner with an untrained answer model is better than pairing it with a CoT-fine-tuned answer model.
  • The data-efficiency result implies that low-data settings, where large CoT corpora are unavailable or too noisy, are exactly where SGFT has the largest relative advantage.

Reading between the lines

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

  • The reported gains likely combine three factors that the paper does not isolate: the SG format, the much stronger GPT-4o teacher that generated the guidance, and the two-model split that offloads answer generation to an untrained model; a controlled comparison that swaps SG data for CoT data of the same size and same teacher would reveal how much of the improvement the format itself contributes.
  • The planner/answerer split resembles a planning-and-execution architecture; if the SG plan is the main driver, one testable extension is to have the answer model execute several alternative SG plans for the same question and vote on the result, which the authors list as future work under self-consistency.
  • Because SG is higher-level, language-agnostic guidance, it may extend beyond math word problems to tasks like multi-hop QA or tool-use planning, where decomposition rather than arithmetic is the bottleneck; the paper's own CommonsenseQA and StrategyQA results are consistent with that direction.
  • A direct scaling prediction follows from Figure 3: accuracy on GSM8K rises with SG training data, so the 3,000-example ceiling is likely not the practical optimum; testing 5,000 or 10,000 SG examples would show whether the gap over CoT widens or saturates.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces Solution Guidance (SG), a reasoning-data format in which a model produces 2–6 high-level solution steps or objectives without performing arithmetic, and Solution-Guidance Fine-Tuning (SGFT), a pipeline that uses GPT-4o to generate SG data for 1,000–3,000 GSM8K questions, fine-tunes a 6–7B small language model with the LISA method to emit such guidance, and then feeds the original question plus the generated guidance to a second, untrained small model that produces the final answer. Across GSM8K, SVAMP, MultiArith, StrategyQA, and CommonsenseQA, the paper reports that SGFT with 3,000 SG examples outperforms CoT fine-tuning with 30,000 examples for ChatGLM3-6B, Qwen2-7B, and Llama2-7B, and it also reports cross-model combinations and ablations over context and few-shot settings during fine-tuning.

Significance. If the central comparison is valid, the practical contribution is meaningful: a roughly 10x reduction in fine-tuning data, single-GPU training, and preservation of the base model's general abilities would make reasoning enhancement more accessible for small models. The paper has clear strengths: it evaluates on five standard benchmarks with three 6–7B families, includes held-out out-of-distribution datasets, provides code and data, and the SG idea of separating problem decomposition from computation is interesting and easy to reproduce. However, the manuscript does not yet establish that the reported gains come from the SG format itself, because the main comparison and the ablation both leave a key control unmeasured; the remaining comments identify the specific missing experiments.

major comments (3)
  1. [§4.2, Table 3] The main comparison in Table 3 does not isolate the SG data format. Every SG row is the SG-fine-tuned generator plus an untuned second model, whereas every CoT baseline is a single fine-tuned model; additionally, the SG training data is produced by GPT-4o and manually cleaned (Section 3.1), while the CoT data source and cleaning are not described at the same level of detail. The observed margin could therefore come from the stronger teacher, from the selection/cleaning of training instances, or from the two-model decomposition rather than from SG as a data format. Please add the missing control: use a CoT-fine-tuned model as the guidance generator with the same untuned answer model, and, if feasible, generate the CoT training set with the same teacher and the same data protocol so that teacher strength and inference architecture are held fixed.
  2. [§4.3, Figure 4] The ablation tests only four combinations: untrained guidance with untrained answering, untrained guidance with a CoT-fine-tuned answer model, SG-fine-tuned guidance with untrained answering, and SG-fine-tuned guidance with a CoT-fine-tuned answer model. The critical symmetric cell, a CoT-fine-tuned model used as the guidance generator with the same untuned answer model, is missing. Without this cell, the improvement attributed to SG could instead be an effect of the two-model decomposition itself. Figure 4 also omits the numeric scores and the model/dataset it refers to; please report those details and add the missing cell.
  3. [§4.1–§4.2, Tables 3–4] No standard deviations, multiple seeds, or significance tests accompany the central accuracy comparisons. The decoding temperature is set to 0, but fine-tuning itself is stochastic, and several of the reported gaps are modest (for example, ChatGLM3-6B on SVAMP: 51.2 vs 47.5, and Qwen2-7B on MultiArith: 61.4 vs 57.2). Please report results over at least a few fine-tuning runs and, if the distributions permit, paired significance tests so that the claim of significant improvement is statistically supported.
minor comments (5)
  1. [Abstract, Table 2] The abstract contains a grammatical error ('this paper introduce') and Table 2 contains '8,00 weeks' in the first example; both should be corrected.
  2. [Conclusion vs §4.2] The conclusion states that 1,000 SG examples outperform 30,000 CoT examples, while Table 3 and Section 4.2 emphasize 3,000 SG examples; please reconcile these numbers and make Figure 3 explicitly support the 1,000-example claim if it is intended.
  3. [§3.1 vs §4.1] Section 3.1 says a subset of 2,000 questions was randomly selected, while Section 4.1 says 1,000, 2,000, and 3,000 questions were used; please clarify whether 2,000 refers to a different selection or is a typo.
  4. [§3.1 and §4.1] The paper uses 'GPT-4' and 'GPT-4o' interchangeably in several places; please use a single naming convention consistently.
  5. [Figure 4 and Table 4] Figure 4 lacks axis labels and numeric values, and Table 4 does not report variance or the exact number of SG examples used; adding these details would make the ablation legible and reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reasoning found; the reported SG advantage is a validity question about missing controls, not a derivation that reduces to its inputs.

full rationale

The paper's central comparison is measured on held-out benchmarks (GSM8K, SVAMP, MultiArith, StrategyQA, CommonsenseQA), and the SG training data is produced by an external teacher, GPT-4o, not by the models whose accuracy is reported. The SG schema is hand-summarized from seven GPT-4o outputs in Section 3.1, but this does not encode test answers, and the response model still must perform the actual computation after receiving the guidance. No fitted parameter is relabeled as a prediction, and no load-bearing result is justified by a self-citation chain; the cited fine-tuning method (LISA) is external prior work. The skeptic's concern, that GPT-4o teacher quality and the two-model inference setup are confounds relative to the CoT baseline, is a threat to the causal attribution of the observed gains, not an instance of circular derivation, because the evaluation remains external and the claimed quantities are not defined in terms of each other.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central claim is empirical and rests on a set of design choices and teacher-model assumptions rather than on a mathematical derivation. The key free choices are the amount of SG data and the SG step template. The most consequential assumptions are about the teacher's quality and the transferability of GSM8K-style guidance to other tasks.

free parameters (5)
  • Number of SG training examples = 1,000 / 2,000 / 3,000
    The paper reports experiments with 1k, 2k, and 3k SG examples and claims 3k outperforms 30k CoT; the amount is a design choice, not derived.
  • Number of few-shot examples during fine-tuning = 3
    The authors limited few-shot examples to 3 due to hardware; this is a hyperparameter.
  • SG step count range = 2 to 6 steps
    Prompt_ii and prompt_v instruct the model to output 2 to 6 steps; this range is chosen by the authors.
  • Number of seed questions for pattern discovery = 7
    They randomly selected 7 GSM8K questions and used GPT-4o to infer a generic solution pattern; M=7 is arbitrary.
  • LISA regularization and layer sampling hyperparameters = not specified
    The paper cites LISA but does not report learning rate, LISA layer sampling schedule, LoRA rank, or epochs; these affect results.
assumptions (3)
  • domain assumption GPT-4o can generate solution guidance that is both accurate and more useful for small models than full chain-of-thought.
    The entire SG data generation depends on the teacher's output quality and format; if the teacher produced noisy or unhelpful plans, the method would fail. Mentioned in Section 3.1.
  • domain assumption GSM8K training questions are a sufficient source domain for guidance that transfers to out-of-distribution datasets such as SVAMP, MultiArith, CommonsenseQA, and StrategyQA.
    The training data comes only from GSM8K, yet the method is claimed to generalize to other reasoning datasets; this assumes transferability.
  • ad hoc to paper CoT fine-tuning with 30,000 samples is an appropriate and representative baseline for the comparison.
    The headline claim of using 3.3% of the data is relative to this specific baseline; the authors do not show that 30k is the saturation point or that the CoT baseline was tuned comparably.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing the Reasoning Capabilities of Small Language Models via Solution Guidance Fine-Tuning." pith.science (2026). https://pith.science/paper/VQEM72DG

@misc{pith2026241209906,
  author       = {Pith},
  title        = {Pith review of: Enhancing the Reasoning Capabilities of Small Language Models via Solution Guidance Fine-Tuning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VQEM72DG}},
  note         = {Machine review of arXiv:2412.09906}
}
read the original abstract

Large language models (LLMs) have demonstrated remarkable performance across a wide range of tasks. Advances in prompt engineering and fine-tuning techniques have further enhanced their ability to address complex reasoning challenges. However, these advanced capabilities are often exclusive to models exceeding 100 billion parameters. Although Chain-of-Thought (CoT) fine-tuning methods have been explored for smaller models (under 10 billion parameters), they typically depend on extensive CoT training data, which can introduce inconsistencies and limit effectiveness in low-data settings. To overcome these limitations, this paper introduce a new reasoning strategy Solution Guidance (SG) and a plug-and-play training paradigm Solution-Guidance Fine-Tuning (SGFT) for enhancing the reasoning capabilities of small language models. SG focuses on problem understanding and decomposition at the semantic and logical levels, rather than specific computations, which can effectively improve the SLMs' generalization and reasoning abilities. With only a small amount of SG training data, SGFT can fine-tune a SLM to produce accurate problem-solving guidances, which can then be flexibly fed to any SLM as prompts, enabling it to generate correct answers directly. Experimental results demonstrate that our method significantly improves the performance of SLMs on various reasoning tasks, enhancing both their practicality and efficiency within resource-constrained environments.

Figures

Figures reproduced from arXiv: 2412.09906 by the authors.

Figure 1
Figure 1. A wrong case of existing CoT reasoning. gation and often generating excessive text, which introduces noise and ultimately affects the accuracy of the final answer. For example, when using CoT reasoning to solve the math problem in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The figure illustrates the complete architecture of our approach. In this diagram, SLM denotes the Small [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Relationship between the amount of training [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparison of the accuracy of four strategies [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. MATATA: Weakly Supervised End-to-End MAthematical Tool-Augmented Reasoning for Tabular Applications

    cs.LG 2024-11 conditional novelty 6.0 of 10

    MATATA uses final-answer-only weak supervision with instruction tuning and KTO preference optimization to train tool-augmented SLM agents that beat or match much larger models on FinQA, TAT-QA, and TabMWP.

  2. Investigating Language Model Capabilities to Represent and Process Formal Knowledge: A Preliminary Study to Assist Ontology Engineering

    cs.AI 2025-09 conditional novelty 5.0 of 10

    On first-order logic reasoning with small language models, the compact CLIF grammar achieves accuracy close to natural language, though natural language remains the best representation.

  3. Error-Aware Curriculum Learning for Biomedical Relation Classification

    cs.CL 2025-07 conditional novelty 5.0 of 10

    A teacher-student pipeline in which GPT-4o diagnoses a student's errors, assigns difficulty scores, and generates remediations, then trains a smaller model by curriculum learning, reports new state-of-the-art F1 on fo...

Reference graph

Works this paper leans on

35 extracted references · 5 canonical work pages · cited by 3 Pith papers

  1. [1]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeff Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Ma teusz Litwin, ...

    arXiv 2020

  2. [2]

    Hyungjoo Chae, Yeonghyeon Kim, Seungone Kim, Kai Tzu iunn Ong, Beong woo Kwak, Moohyeon Kim, Seonghwan Kim, Taeyoon Kwon, Jiwan Chung, Youngjae Yu, and Jinyoung Yeo. 2024. https://api.semanticscholar.org/CorpusID:268876471 Language models as compilers: Simulating pseudocode execution improves algorithmic reasoning in language models . ArXiv, abs/2404.02575

    arXiv 2024

  3. [3]

    Gawon Choi and Hyemin Ahn. 2024. https://api.semanticscholar.org/CorpusID:268987756 Can only llms do reasoning?: Potential of small language models in task planning . ArXiv, abs/2404.03891

    work page Pith review arXiv 2024

  4. [4]

    Yunfei Chu, Jin Xu, Qian Yang, Haojie Wei, Xipin Wei, Zhifang Guo, Yichong Leng, Yuanjun Lv, Jinzheng He, Junyang Lin, Chang Zhou, and Jingren Zhou. 2024. https://api.semanticscholar.org/CorpusID:271213498 Qwen2-audio technical report . ArXiv, abs/2407.10759

    arXiv 2024

  5. [5]

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. https://api.semanticscholar.org/CorpusID:239998651 Training verifiers to solve math word problems . ArXiv, abs/2110.14168

    arXiv 2021

  6. [6]

    Shizhe Diao, Pengcheng Wang, Yong Lin, Xiang Liu, and Tong Zhang. 2023. https://api.semanticscholar.org/CorpusID:257102707 Active prompting with chain-of-thought for large language models . In Annual Meeting of the Association for Computational Linguistics

    work page 2023

  7. [7]

    Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Zhiyong Wu, Baobao Chang, Xu Sun, Jingjing Xu, Lei Li, and Zhifang Sui. 2023. https://api.semanticscholar.org/CorpusID:263886074 A survey for in-context learning . ArXiv, abs/2301.00234

    arXiv 2023

  8. [8]

    Yao Fu, Hao-Chun Peng, Litu Ou, Ashish Sabharwal, and Tushar Khot. 2023. https://api.semanticscholar.org/CorpusID:256390607 Specializing smaller language models towards multi-step reasoning . ArXiv, abs/2301.12726

    arXiv 2023

Show all 35 references
  1. [9]

    Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. 2021. https://api.semanticscholar.org/CorpusID:230799347 Did aristotle use a laptop? a question answering benchmark with implicit reasoning strategies . Transactions of the Association for Compu...

    2021

  2. [10]

    Namgyu Ho, Laura Schmid, and Se-Young Yun. 2022. https://api.semanticscholar.org/CorpusID:254877399 Large language models are reasoning teachers . In Annual Meeting of the Association for Computational Linguistics

    2022

  3. [11]

    Edward Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, and Weizhu Chen

    J. Edward Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, and Weizhu Chen. 2021. https://api.semanticscholar.org/CorpusID:235458009 Lora: Low-rank adaptation of large language models . ArXiv, abs/2106.09685

    2021 arXiv

  4. [12]

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. https://api.semanticscholar.org/CorpusID:249017743 Large language models are zero-shot reasoners . ArXiv, abs/2205.11916

    2022 arXiv

  5. [13]

    Tamera Lanham, Anna Chen, Ansh Radhakrishnan, Benoit Steiner, Carson E. Denison, Danny Hernandez, Dustin Li, Esin Durmus, Evan Hubinger, John Kernion, Kamil.e Lukovsiut.e, Karina Nguyen, Newton Cheng, Nicholas Joseph, Nicholas Schiefer, Oliver Rausch, Robin Larson, Sam McCandl...

    2023 arXiv

  6. [14]

    Shiyang Li, Jianshu Chen, Yelong Shen, Zhiyu Chen, Xinlu Zhang, Zekun Li, Hong Wang, Jingu Qian, Baolin Peng, Yi Mao, Wenhu Chen, and Xifeng Yan. 2022. https://api.semanticscholar.org/CorpusID:252873123 Explanations from large language models make small reasoners better . ArXi...

    2022 arXiv

  7. [15]

    Pan Lu, Liang Qiu, Wenhao Yu, Sean Welleck, and Kai-Wei Chang. 2022. https://api.semanticscholar.org/CorpusID:254877175 A survey of deep learning for mathematical reasoning . ArXiv, abs/2212.10535

    2022 arXiv

  8. [16]

    Lucie Charlotte Magister, Jonathan Mallinson, Jakub Adamek, Eric Malmi, and Aliaksei Severyn. 2022. https://api.semanticscholar.org/CorpusID:254823156 Teaching small language models to reason . ArXiv, abs/2212.08410

    2022 arXiv

  9. [17]

    Swaroop Mishra, Matthew Finlayson, Pan Lu, Leonard Tang, Sean Welleck, Chitta Baral, Tanmay Rajpurohit, Oyvind Tafjord, Ashish Sabharwal, Peter Clark, and A. Kalyan. 2022. https://api.semanticscholar.org/CorpusID:253237047 Lila: A unified benchmark for mathematical reasoning ....

    2022 arXiv

  10. [18]

    Rui Pan, Xiang Liu, Shizhe Diao, Renjie Pi, Jipeng Zhang, Chi Han, and Tong Zhang. 2024. https://api.semanticscholar.org/CorpusID:268691349 Lisa: Layerwise importance sampling for memory-efficient large language model fine-tuning . ArXiv, abs/2403.17919

    2024 arXiv

  11. [19]

    Bhattamishra, and Navin Goyal

    Arkil Patel, S. Bhattamishra, and Navin Goyal. 2021. https://api.semanticscholar.org/CorpusID:232223322 Are nlp models really able to solve simple math word problems? In North American Chapter of the Association for Computational Linguistics

    2021

  12. [20]

    Shuofei Qiao, Yixin Ou, Ningyu Zhang, Xiang Chen, Yunzhi Yao, Shumin Deng, Chuanqi Tan, Fei Huang, and Huajun Chen. 2022. https://api.semanticscholar.org/CorpusID:254854219 Reasoning with language model prompting: A survey . ArXiv, abs/2212.09597

    2022 arXiv

  13. [21]

    Subhro Roy and Dan Roth. 2016. https://api.semanticscholar.org/CorpusID:560565 Solving general arithmetic word problems . ArXiv, abs/1608.01413

    2016 arXiv

  14. [22]

    Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. 2019. https://api.semanticscholar.org/CorpusID:53296520 Commonsenseqa: A question answering challenge targeting commonsense knowledge . ArXiv, abs/1811.00937

    2019 arXiv

  15. [23]

    Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Daniel M

    Hugo Touvron, Louis Martin, Kevin R. Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Daniel M. Bikel, Lukas Blecher, Cristian Cant \'o n Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes,...

    2023 arXiv

  16. [24]

    Miles Turpin, Julian Michael, Ethan Perez, and Sam Bowman. 2023. https://api.semanticscholar.org/CorpusID:258556812 Language models don't always say what they think: Unfaithful explanations in chain-of-thought prompting . ArXiv, abs/2305.04388

    2023 arXiv

  17. [25]

    Wang, Qiushi Sun, Nuo Chen, Xiang Lorraine Li, and Ming Gao

    J. Wang, Qiushi Sun, Nuo Chen, Xiang Lorraine Li, and Ming Gao. 2023. https://api.semanticscholar.org/CorpusID:259138909 Boosting language models reasoning with chain-of-knowledge prompting . ArXiv, abs/2306.06427

    2023 arXiv

  18. [26]

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Huai hsin Chi, and Denny Zhou. 2022. https://api.semanticscholar.org/CorpusID:247595263 Self-consistency improves chain of thought reasoning in language models . ArXiv, abs/2203.11171

    2022 arXiv

  19. [27]

    Xia, Quoc Le, and Denny Zhou

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Huai hsin Chi, F. Xia, Quoc Le, and Denny Zhou. 2022. https://api.semanticscholar.org/CorpusID:246411621 Chain of thought prompting elicits reasoning in large language models . ArXiv, abs/2201.11903

    2022 arXiv

  20. [28]

    Cambria, Xiaodong Liu, Jianfeng Gao, and Furu Wei

    Zonglin Yang, Li Dong, Xinya Du, Hao Cheng, E. Cambria, Xiaodong Liu, Jianfeng Gao, and Furu Wei. 2022. https://api.semanticscholar.org/CorpusID:254926851 Language models as inductive reasoners . ArXiv, abs/2212.10923

    2022 arXiv

  21. [29]

    Griffiths, Yuan Cao, and Karthik Narasimhan

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. https://api.semanticscholar.org/CorpusID:258762525 Tree of thoughts: Deliberate problem solving with large language models . ArXiv, abs/2305.10601

    2023 arXiv

  22. [30]

    Team Glm Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Diego Rojas, Guanyu Feng, Hanlin Zhao, Hanyu Lai, Hao Yu, Hongning Wang, Jiadai Sun, Jiajie Zhang, Jiale Cheng, Jiayi Gui, Jie Tang, Jing Zhang, Juanzi Li, Lei Zhao, Lindong Wu, Lucen Zhong, Ming yue Liu, Minlie H...

    2024 arXiv

  23. [31]

    Zhuosheng Zhang, Aston Zhang, Mu Li, and Alexander J. Smola. 2022. https://api.semanticscholar.org/CorpusID:252762275 Automatic chain of thought prompting in large language models . ArXiv, abs/2210.03493

    2022 arXiv

  24. [32]

    Qihuang Zhong, Kang Wang, Ziyang Xu, Juhua Liu, Liang Ding, Bo Du, and Dacheng Tao. 2024. https://api.semanticscholar.org/CorpusID:269448917 Achieving >97\ ArXiv, abs/2404.14963

    2024 arXiv

  25. [33]

    Xunyu Zhu, Jian Li, Yong Liu, Can Ma, and Weiping Wang. 2023. https://api.semanticscholar.org/CorpusID:260900101 A survey on model compression for large language models . ArXiv, abs/2308.07633

    2023 arXiv

  26. [34]

    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...

  27. [35]

    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 11, 2026 · model on record in the stance chip above.