Resource-Aware Neuro-Symbolic Reasoning for Local Small Language Models

Abel Alvarez; Carlos Ram\'irez Ovalle

arxiv: 2606.27281 · v1 · pith:RIFFPV42new · submitted 2026-06-25 · 💻 cs.LO

Resource-Aware Neuro-Symbolic Reasoning for Local Small Language Models

Carlos Ram\'irez Ovalle , Abel Alvarez This is my paper

Pith reviewed 2026-06-26 01:56 UTC · model grok-4.3

classification 💻 cs.LO

keywords neuro-symbolic reasoningsmall language modelsverifiable formalizationfinite-domain logicresource-aware inferencelogical deductionlocal inference

0 comments

The pith

A neuro-symbolic pipeline lets small local language models reach over 0.98 accuracy on logic tasks with one model call instead of five.

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

The paper tests whether converting problems into typed finite-domain rules and constraints, then verifying them symbolically before deterministic solving, can improve accuracy and efficiency for small language models on local hardware. On 120 generated pure-precedence problems the approach reaches 0.983 accuracy versus 0.700 for serial self-consistency, and on a 120-instance BBH-derived logical-deduction set it reaches 0.933 versus 0.283, all while using a single model call. The VFR-LLM pipeline is evaluated with Qwen3-4B on Apple Silicon, with robustness checks on other models, and the gains hold against cost-aware adaptive baselines. The work shows the method is bounded to finite-domain logic but saves calls and latency for those tasks.

Core claim

The Verifiable Formalization and Repair pipeline (VFR-LLM) has the model translate a problem into typed finite-domain rules and constraints, after which a symbolic layer checks traceability and consistency and a deterministic solver performs the reasoning. On local hardware this yields 0.983 accuracy on pure-precedence problems and 0.933 on extended logical-deduction instances, each using one model call versus five for serial self-consistency, with the gap persisting against stronger baselines.

What carries the argument

The VFR-LLM pipeline, in which the language model generates typed finite-domain rules and constraints that a symbolic verification layer confirms before a deterministic solver computes the answer.

If this is right

Finite-domain logic tasks can be solved with one model call rather than repeated sampling while raising accuracy from 0.700 to 0.983 on precedence problems.
The same single-call advantage appears on BBH-derived deduction instances, moving accuracy from 0.283 to 0.933.
Performance remains model-dependent, with some models failing to produce usable typed constraints.
Token usage and serial latency drop for tasks where the symbolic solver can replace further sampling.

Where Pith is reading between the lines

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

The method may extend to other structured domains that admit finite-domain encodings and exact solvers.
Hybrid verification could reduce the impact of occasional LLM errors on downstream symbolic steps in local deployments.
Further scaling the constraint complexity would test whether verification overhead remains negligible relative to sampling cost.

Load-bearing premise

The language model produces typed finite-domain rules and constraints that the symbolic layer can confirm as consistent and traceable without translation errors that would invalidate the solver output.

What would settle it

A collection of problems where the generated rules contain subtle inconsistencies the verifier accepts but that cause the solver to return demonstrably incorrect answers.

read the original abstract

Small language models can run locally on consumer hardware, but structured reasoning often pushes users toward repeated sampling or larger models. This article studies a bounded neuro-symbolic alternative for local inference: a model translates a problem into typed finite-domain rules and constraints, a symbolic layer checks traceability and consistency, and a deterministic solver performs the reasoning step. The resulting Verifiable Formalization and Repair pipeline (VFR-LLM) tests when symbolic verification can replace repeated sampling without weakening accuracy. We evaluate it through LM Studio on Apple Silicon, using Qwen3-4B-2507 as the primary model, with Phi-4-mini-reasoning and Gemma-3n-E4B as robustness checks. On 120 generated pure-precedence problems, Qwen VFR-LLM achieves 0.983 accuracy, versus 0.700 for serial self-consistency using one model call instead of five. On a 120-instance BBH-derived extended logical-deduction subset, it reaches 0.933 versus 0.283. The advantage persists against a stronger cost-aware adaptive self-consistency baseline, which lowers sampling cost but not the single-call accuracy gap. Gemma reproduces the same model-dependent boundaries and Phi is negative on typed constraints. The contribution is bounded: finite-domain logic can replace repeated local sampling for some structured tasks, saving model calls and serial latency, with stratum-dependent token savings.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

VFR-LLM shows a single-call accuracy edge on small local models for precedence and deduction tasks, but the symbolic layer only verifies internal consistency of the formalization, not whether it matches the original problem semantics.

read the letter

The main point is that this pipeline lets a 4B model reach 0.983 accuracy on 120 generated precedence problems and 0.933 on a 120-instance BBH logical deduction subset, using one call instead of the five needed for serial self-consistency. It works by having the model output typed finite-domain rules and constraints, running a symbolic check for traceability and consistency, then handing off to a deterministic solver. The advantage holds against an adaptive self-consistency baseline too, and they show it on Apple Silicon hardware with Qwen as the main model plus Phi and Gemma checks.

What the paper does is give a concrete, bounded demonstration that this neuro-symbolic handoff can reduce sampling cost and latency for certain structured tasks on consumer hardware. The model comparisons and the stratum-dependent token savings are reported directly, and the contribution statement stays modest.

The soft spot is exactly the one the stress test flags. The verification step confirms that the generated rules are consistent with each other and traceable, but it does not check whether those rules correctly encode the natural-language problem. A translation error that produces a consistent but non-equivalent formalization will still pass verification and produce a wrong final answer. The reported gaps therefore depend on the untested assumption that such semantic mismatches are rare on these tasks. The datasets are generated or derived subsets, so baseline fairness also needs checking in the full methods.

No circular definitions or free parameters stand out in the reported results. This is for engineers and researchers focused on practical local inference for logic-style problems. A reader who wants numbers on single-call neuro-symbolic tradeoffs will find usable data here.

It deserves peer review because the accuracy differences are large enough to be worth examining, even with the semantic-fidelity question left open.

Referee Report

1 major / 2 minor

Summary. The manuscript presents the Verifiable Formalization and Repair (VFR-LLM) pipeline for neuro-symbolic reasoning with small local language models. An LLM translates a problem into typed finite-domain rules and constraints; a symbolic layer verifies traceability and consistency; a deterministic solver computes the answer. On 120 generated pure-precedence problems, Qwen3-4B VFR-LLM reaches 0.983 accuracy (vs. 0.700 for serial self-consistency); on a 120-instance BBH-derived logical-deduction subset, it reaches 0.933 (vs. 0.283). The approach uses one model call instead of five and is tested on Apple Silicon via LM Studio with robustness checks on other models.

Significance. If the central assumption holds, the work shows that for certain structured reasoning tasks, symbolic verification can substitute for repeated sampling in local small-model settings, reducing call count and serial latency while maintaining or improving accuracy. The stratum-dependent token savings and model-specific boundaries (positive for Qwen, negative for Phi) provide concrete guidance on when the method applies.

major comments (1)

[Abstract / pipeline description] The reported accuracy figures rest on the assumption that LLM-generated formalizations are semantically equivalent to the original natural-language problems. However, the symbolic verification layer only confirms internal consistency and traceability of the generated rules/constraints; it does not check whether those rules correctly capture the problem semantics (e.g., correct precedence relations). A translation error producing a consistent but non-equivalent formalization would pass verification and produce an incorrect solver output that is counted as a success, directly undermining the claimed accuracy gaps versus self-consistency baselines.

minor comments (2)

[Abstract] No details are provided on how the 120 generated pure-precedence problems or the BBH-derived subset were constructed, nor on error analysis or failure modes.
[Abstract] The comparison to the 'stronger cost-aware adaptive self-consistency baseline' lacks quantitative details on its sampling cost and accuracy in the provided text.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed comment on the distinction between internal verification and semantic equivalence. We address the point directly below and will revise the manuscript to clarify the evaluation methodology.

read point-by-point responses

Referee: [Abstract / pipeline description] The reported accuracy figures rest on the assumption that LLM-generated formalizations are semantically equivalent to the original natural-language problems. However, the symbolic verification layer only confirms internal consistency and traceability of the generated rules/constraints; it does not check whether those rules correctly capture the problem semantics (e.g., correct precedence relations). A translation error producing a consistent but non-equivalent formalization would pass verification and produce an incorrect solver output that is counted as a success, directly undermining the claimed accuracy gaps versus self-consistency baselines.

Authors: We agree that the symbolic verification layer confirms only internal consistency and traceability, not semantic equivalence to the input problem. The reported accuracies, however, are strictly end-to-end: they measure the fraction of cases in which the solver output matches the ground-truth answer to the original natural-language problem. A non-equivalent formalization will, except in the rare case where the semantic error leaves the queried answer unchanged, produce a solver result that fails to match ground truth and is therefore scored as an error. The high accuracies (0.983 on precedence, 0.933 on logical deduction) therefore indicate that, for the tested tasks and models, the LLM formalizations were semantically faithful whenever verification passed. This evaluation protocol is identical to that used for the self-consistency baselines, so the accuracy comparison remains valid. We will revise the abstract and pipeline section to state explicitly that semantic correctness is validated indirectly through ground-truth agreement rather than by the verifier, and we will add a short discussion of the implicit nature of this check. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical accuracy claims rest on direct measurement, not self-referential definitions or fitted inputs

full rationale

The paper describes a neuro-symbolic pipeline (LLM formalization + symbolic consistency check + deterministic solver) and reports empirical accuracies (0.983 / 0.933) on 120 generated precedence problems and a BBH-derived subset versus self-consistency baselines. No equations, parameters, or derivations are presented that reduce the reported accuracies to inputs by construction. No self-citations, uniqueness theorems, or ansatzes appear in the provided text. The evaluation is a straightforward comparison of end-to-end accuracy on fixed test sets; the central claims therefore remain independent of the pipeline's internal mechanics and receive score 0.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the pipeline implicitly assumes reliable LLM formalization and finite-domain solvability without further specification.

pith-pipeline@v0.9.1-grok · 5778 in / 1163 out tokens · 25473 ms · 2026-06-26T01:56:11.555884+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

22 extracted references · 7 canonical work pages

[1]

Large language models (LLMs): survey, technical frameworks, and future challenges

Pranjal Kumar. Large language models (LLMs): survey, technical frameworks, and future challenges. Artificial Intelligence Review, 57:260, 2024. doi: 10.1007/s10462-024-10888-y. URL https: //link.springer.com/article/10.1007/s10462-024-10888-y

work page doi:10.1007/s10462-024-10888-y 2024
[2]

Empowering large language models to edge intelligence: A survey of edge efficient LLMs and techniques.Computer Science Review, 57:100755, 2025

Rui Wang, Zhiyong Gao, Liuyang Zhang, Shuaibing Yue, and Ziyi Gao. Empowering large language models to edge intelligence: A survey of edge efficient LLMs and techniques.Computer Science Review, 57:100755, 2025. doi: 10.1016/j.cosrev.2025.100755. URL https://doi.org/10. 1016/j.cosrev.2025.100755

work page doi:10.1016/j.cosrev.2025.100755 2025
[3]

State of the art and future directions of small language models: A systematic review.Big Data and Cognitive Computing, 9(7):189, 2025

Flavio Corradini, Matteo Leonesi, and Marco Piangerelli. State of the art and future directions of small language models: A systematic review.Big Data and Cognitive Computing, 9(7):189, 2025. doi: 10.3390/bdcc9070189. URLhttps://doi.org/10.3390/bdcc9070189

work page doi:10.3390/bdcc9070189 2025
[4]

SmallThinker: A family of efficient large language models natively trained for local deployment.arXiv preprint arXiv:2507.20984, 2025

Yixin Song, Zhenliang Xue, Dongliang Wei, Feiyang Chen, Jianxiang Gao, Junchen Liu, Hangyu Liang, Guangshuo Qin, Chengrong Tian, Bo Wen, Longyu Zhao, Xinrui Zheng, Zeyu Mi, and Haibo Chen. SmallThinker: A family of efficient large language models natively trained for local deployment.arXiv preprint arXiv:2507.20984, 2025. URLhttps://arxiv.org/abs/2507.20984

arXiv 2025
[5]

Neuro-symbolic approaches in artificial intelligence.National Science Review, 9(6), 2022

Pascal Hitzler. Neuro-symbolic approaches in artificial intelligence.National Science Review, 9(6), 2022

2022
[6]

Self-consistency improves chain of thought reasoning in language models

Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models. In International Conference on Learning Representations, 2023

2023
[7]

Let’s sample step by step: Adaptive- consistency for efficient reasoning and coding with LLMs

Pranjal Aggarwal, Aman Madaan, Yiming Yang, and Mausam. Let’s sample step by step: Adaptive- consistency for efficient reasoning and coding with LLMs. InProceedings of the 2023 Con- ference on Empirical Methods in Natural Language Processing, pages 12375–12396, Singapore,

2023
[8]

Let ' s Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLM s

Association for Computational Linguistics. doi: 10.18653/v1/2023.emnlp-main.761. URL https://aclanthology.org/2023.emnlp-main.761/

work page doi:10.18653/v1/2023.emnlp-main.761 2023
[9]

Make every penny count: Difficulty-adaptive self-consistency for cost-efficient reasoning

Xinglin Wang, Shaoxiong Feng, Yiwei Li, Peiwen Yuan, Yueqi Zhang, Chuyi Tan, Boyuan Pan, Yao Hu, and Kan Li. Make every penny count: Difficulty-adaptive self-consistency for cost-efficient reasoning. InFindings of the Association for Computational Linguistics: NAACL 2025, pages 6919– 6932, Albuquerque, New Mexico, 2025. Association for Computational Lingu...

2025
[10]

What you always wanted to know about datalog (and never dared to ask).IEEE Transactions on Knowledge and Data Engineering, 1(1):146–166, 1989

Stefano Ceri, Georg Gottlob, and Letizia Tanca. What you always wanted to know about datalog (and never dared to ask).IEEE Transactions on Knowledge and Data Engineering, 1(1):146–166, 1989

1989
[11]

Z3: An efficient SMT solver

Leonardo De Moura and Nikolaj Bjørner. Z3: An efficient SMT solver. InTools and Algorithms for the Construction and Analysis of Systems, pages 337–340. Springer, 2008

2008
[12]

Augmented language models: a survey.Transactions on Machine Learning Research, 2023

Grégoire Mialon, Roberto Dessì, Maria Lomeli, Christoforos Nalmpantis, Ram Pasunuru, Roberta Raileanu, Baptiste Rozière, Timo Schick, Jane Dwivedi-Yu, Asli Celikyilmaz, et al. Augmented language models: a survey.Transactions on Machine Learning Research, 2023. URL https: //openreview.net/forum?id=jh7wH2AzKK. 23

2023
[13]

ReAct: Synergizing reasoning and acting in language models

Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models. InInternational Conference on Learning Representations, 2023

2023
[14]

Toolformer: Language models can teach them- selves to use tools

Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Language models can teach them- selves to use tools. InAdvances in Neural Information Processing Systems, volume 36. Curran Associates, Inc., 2023

2023
[15]

Pal: Program-aided language models

Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu, Yiming Yang, Jamie Callan, and Gra- ham Neubig. Pal: Program-aided language models. InProceedings of the 40th International Conference on Machine Learning, volume 202 ofProceedings of Machine Learning Research, pages 10764–10799. PMLR, 2023. URLhttps://proceedings.mlr.press/v202/gao23f.html

2023
[16]

Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W. Cohen. Program of thoughts prompt- ing: Disentangling computation from reasoning for numerical reasoning tasks.Transactions on Machine Learning Research, 2023. URL https://mlanthology.org/tmlr/2023/ chen2023tmlr-program/

2023
[17]

Logic-lm: Empowering large language models with symbolic solvers for faithful logical reasoning

Liangming Pan, Alon Albalak, Xinyi Wang, and William Wang. Logic-lm: Empowering large language models with symbolic solvers for faithful logical reasoning. InFindings of the Association for Com- putational Linguistics: EMNLP 2023, pages 3806–3824. Association for Computational Linguistics,

2023
[18]

Logic- LM : Empowering Large Language Models with Symbolic Solvers for Faithful Logical Reasoning

doi: 10.18653/v1/2023.findings-emnlp.248. URL https://aclanthology.org/2023. findings-emnlp.248/

work page doi:10.18653/v1/2023.findings-emnlp.248 2023
[19]

2023 , issue_date =

Ziyang Li, Jiani Huang, and Mayur Naik. Scallop: A language for neurosymbolic programming. Proceedings of the ACM on Programming Languages, 7(PLDI):1463–1487, 2023. doi: 10.1145/ 3591280. URLhttps://doi.org/10.1145/3591280

work page doi:10.1145/3591280 2023
[20]

TinyNS: Platform-aware neurosymbolic auto tiny machine learning.ACM Transactions on Embedded Computing Systems, 23(3):43, 2024

Swapnil Sayan Saha, Sandeep Singh Sandha, Mohit Aggarwal, Brian Wang, Liying Han, Julian De Gortari Briseno, and Mani Srivastava. TinyNS: Platform-aware neurosymbolic auto tiny machine learning.ACM Transactions on Embedded Computing Systems, 23(3):43, 2024. doi: 10.1145/3603171. URLhttps://doi.org/10.1145/3603171

work page doi:10.1145/3603171 2024
[21]

Le, Ed H

Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc V . Le, Ed H. Chi, Denny Zhou, and Jason Wei. Challenging BIG-Bench tasks and whether chain-of-thought can solve them. InFindings of the Association for Computational Linguistics: ACL 2023, pages 13003–13051. Association for Computational...

2023
[22]

Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, et al

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Adam R. Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, et al. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models.Transactions on Machine Learning Research, 2023. ISSN 2835-8856. URL https://openreview.net/foru...

2023

[1] [1]

Large language models (LLMs): survey, technical frameworks, and future challenges

Pranjal Kumar. Large language models (LLMs): survey, technical frameworks, and future challenges. Artificial Intelligence Review, 57:260, 2024. doi: 10.1007/s10462-024-10888-y. URL https: //link.springer.com/article/10.1007/s10462-024-10888-y

work page doi:10.1007/s10462-024-10888-y 2024

[2] [2]

Empowering large language models to edge intelligence: A survey of edge efficient LLMs and techniques.Computer Science Review, 57:100755, 2025

Rui Wang, Zhiyong Gao, Liuyang Zhang, Shuaibing Yue, and Ziyi Gao. Empowering large language models to edge intelligence: A survey of edge efficient LLMs and techniques.Computer Science Review, 57:100755, 2025. doi: 10.1016/j.cosrev.2025.100755. URL https://doi.org/10. 1016/j.cosrev.2025.100755

work page doi:10.1016/j.cosrev.2025.100755 2025

[3] [3]

State of the art and future directions of small language models: A systematic review.Big Data and Cognitive Computing, 9(7):189, 2025

Flavio Corradini, Matteo Leonesi, and Marco Piangerelli. State of the art and future directions of small language models: A systematic review.Big Data and Cognitive Computing, 9(7):189, 2025. doi: 10.3390/bdcc9070189. URLhttps://doi.org/10.3390/bdcc9070189

work page doi:10.3390/bdcc9070189 2025

[4] [4]

SmallThinker: A family of efficient large language models natively trained for local deployment.arXiv preprint arXiv:2507.20984, 2025

Yixin Song, Zhenliang Xue, Dongliang Wei, Feiyang Chen, Jianxiang Gao, Junchen Liu, Hangyu Liang, Guangshuo Qin, Chengrong Tian, Bo Wen, Longyu Zhao, Xinrui Zheng, Zeyu Mi, and Haibo Chen. SmallThinker: A family of efficient large language models natively trained for local deployment.arXiv preprint arXiv:2507.20984, 2025. URLhttps://arxiv.org/abs/2507.20984

arXiv 2025

[5] [5]

Neuro-symbolic approaches in artificial intelligence.National Science Review, 9(6), 2022

Pascal Hitzler. Neuro-symbolic approaches in artificial intelligence.National Science Review, 9(6), 2022

2022

[6] [6]

Self-consistency improves chain of thought reasoning in language models

Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models. In International Conference on Learning Representations, 2023

2023

[7] [7]

Let’s sample step by step: Adaptive- consistency for efficient reasoning and coding with LLMs

Pranjal Aggarwal, Aman Madaan, Yiming Yang, and Mausam. Let’s sample step by step: Adaptive- consistency for efficient reasoning and coding with LLMs. InProceedings of the 2023 Con- ference on Empirical Methods in Natural Language Processing, pages 12375–12396, Singapore,

2023

[8] [8]

Let ' s Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLM s

Association for Computational Linguistics. doi: 10.18653/v1/2023.emnlp-main.761. URL https://aclanthology.org/2023.emnlp-main.761/

work page doi:10.18653/v1/2023.emnlp-main.761 2023

[9] [9]

Make every penny count: Difficulty-adaptive self-consistency for cost-efficient reasoning

Xinglin Wang, Shaoxiong Feng, Yiwei Li, Peiwen Yuan, Yueqi Zhang, Chuyi Tan, Boyuan Pan, Yao Hu, and Kan Li. Make every penny count: Difficulty-adaptive self-consistency for cost-efficient reasoning. InFindings of the Association for Computational Linguistics: NAACL 2025, pages 6919– 6932, Albuquerque, New Mexico, 2025. Association for Computational Lingu...

2025

[10] [10]

What you always wanted to know about datalog (and never dared to ask).IEEE Transactions on Knowledge and Data Engineering, 1(1):146–166, 1989

Stefano Ceri, Georg Gottlob, and Letizia Tanca. What you always wanted to know about datalog (and never dared to ask).IEEE Transactions on Knowledge and Data Engineering, 1(1):146–166, 1989

1989

[11] [11]

Z3: An efficient SMT solver

Leonardo De Moura and Nikolaj Bjørner. Z3: An efficient SMT solver. InTools and Algorithms for the Construction and Analysis of Systems, pages 337–340. Springer, 2008

2008

[12] [12]

Augmented language models: a survey.Transactions on Machine Learning Research, 2023

Grégoire Mialon, Roberto Dessì, Maria Lomeli, Christoforos Nalmpantis, Ram Pasunuru, Roberta Raileanu, Baptiste Rozière, Timo Schick, Jane Dwivedi-Yu, Asli Celikyilmaz, et al. Augmented language models: a survey.Transactions on Machine Learning Research, 2023. URL https: //openreview.net/forum?id=jh7wH2AzKK. 23

2023

[13] [13]

ReAct: Synergizing reasoning and acting in language models

Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models. InInternational Conference on Learning Representations, 2023

2023

[14] [14]

Toolformer: Language models can teach them- selves to use tools

Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Language models can teach them- selves to use tools. InAdvances in Neural Information Processing Systems, volume 36. Curran Associates, Inc., 2023

2023

[15] [15]

Pal: Program-aided language models

Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu, Yiming Yang, Jamie Callan, and Gra- ham Neubig. Pal: Program-aided language models. InProceedings of the 40th International Conference on Machine Learning, volume 202 ofProceedings of Machine Learning Research, pages 10764–10799. PMLR, 2023. URLhttps://proceedings.mlr.press/v202/gao23f.html

2023

[16] [16]

Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W. Cohen. Program of thoughts prompt- ing: Disentangling computation from reasoning for numerical reasoning tasks.Transactions on Machine Learning Research, 2023. URL https://mlanthology.org/tmlr/2023/ chen2023tmlr-program/

2023

[17] [17]

Logic-lm: Empowering large language models with symbolic solvers for faithful logical reasoning

Liangming Pan, Alon Albalak, Xinyi Wang, and William Wang. Logic-lm: Empowering large language models with symbolic solvers for faithful logical reasoning. InFindings of the Association for Com- putational Linguistics: EMNLP 2023, pages 3806–3824. Association for Computational Linguistics,

2023

[18] [18]

Logic- LM : Empowering Large Language Models with Symbolic Solvers for Faithful Logical Reasoning

doi: 10.18653/v1/2023.findings-emnlp.248. URL https://aclanthology.org/2023. findings-emnlp.248/

work page doi:10.18653/v1/2023.findings-emnlp.248 2023

[19] [19]

2023 , issue_date =

Ziyang Li, Jiani Huang, and Mayur Naik. Scallop: A language for neurosymbolic programming. Proceedings of the ACM on Programming Languages, 7(PLDI):1463–1487, 2023. doi: 10.1145/ 3591280. URLhttps://doi.org/10.1145/3591280

work page doi:10.1145/3591280 2023

[20] [20]

TinyNS: Platform-aware neurosymbolic auto tiny machine learning.ACM Transactions on Embedded Computing Systems, 23(3):43, 2024

Swapnil Sayan Saha, Sandeep Singh Sandha, Mohit Aggarwal, Brian Wang, Liying Han, Julian De Gortari Briseno, and Mani Srivastava. TinyNS: Platform-aware neurosymbolic auto tiny machine learning.ACM Transactions on Embedded Computing Systems, 23(3):43, 2024. doi: 10.1145/3603171. URLhttps://doi.org/10.1145/3603171

work page doi:10.1145/3603171 2024

[21] [21]

Le, Ed H

Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc V . Le, Ed H. Chi, Denny Zhou, and Jason Wei. Challenging BIG-Bench tasks and whether chain-of-thought can solve them. InFindings of the Association for Computational Linguistics: ACL 2023, pages 13003–13051. Association for Computational...

2023

[22] [22]

Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, et al

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Adam R. Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, et al. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models.Transactions on Machine Learning Research, 2023. ISSN 2835-8856. URL https://openreview.net/foru...

2023