Pith. sign in

REVIEW 4 major objections 3 minor 42 references

Test-Time Scaling for CAD Generation via Verifier-Free Consensus Selection

T0 review · 4 major / 3 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 Geometric consensus among generated CAD models selects better shapes than a vision-language verifier from the same candidate pool.

desk verdict Agreement-based selection beats a VLM verifier on the same pools, but the mechanism's failure mode under systematic error is acknowledged, not tested. read the letter →

arxiv 2608.09706 v1 pith:D4CLBRT7 submitted 2026-08-10 cs.CE cs.LG

Aaron Haag , Altay Kaçan , Bertram Fuchs , Oliver Lohse This is my paper · ORCID
classification cs.CEcs.LG
keywords test-timescalingtext-to-CADgenerationconsensusselectionminimumBayesriskdecodingparametricCADChamferdistanceEulercharacteristicverifier-free
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 asks whether the candidate pool itself is enough to select a good parametric CAD model when no ground truth and no verifier is available. It proposes consensus selection: sample N CAD programs, compile them to 3D models, and return the candidate with the lowest average distance to all other candidates. On the exact candidate pools of a state-of-the-art verifier-based CAD agent, geometric consensus improves all three geometric metrics over that agent's verifier, while a topological consensus variant matches the verifier on topology. Across every tested LLM and prompt variant, geometric consensus also beats random selection from the same pool. This makes agreement a cheaper, training-free alternative to vision-language judges for choosing among CAD candidates.

What carries the argument

The load-bearing object is the consensus selection rule, Equation (1), which turns a pool of compiled CAD models into a single pick without ground truth or a verifier. The rule is instantiated with two distances: the symmetric Chamfer distance $d_{CD}$ between uniformly sampled, ICP-aligned surface point clouds, which yields the geometric medoid of the pool; and a 0-1 distance $d_\chi$ on the Euler characteristic of watertight meshes, which yields a majority vote over topology classes, with geometric consensus breaking ties. The choice of distance determines which evaluation metric improves, and the rule is training-free, requiring only a compiler plus $O(|V|^2)$ pairwise distance evaluations.

What would settle it

A controlled experiment that injects a shared systematic error into most candidates—for example, all but one generated model omit a slot that the prompt requires—would settle the claim: if geometric consensus then picks the erroneous majority and its Chamfer distance to the ground truth is worse than random selection or the oracle on those prompts, the central assumption fails; if it still tracks correctness, the claim survives. Concretely, compute per-prompt the rank correlation between a candidate's mean distance to the pool and its distance to the ground-truth model; prompts with strong negative correlation would be direct counterexamples.

Watch

Extended reading notes

Core claim

The paper's central claim is that for text-to-CAD generation, selection by agreement over compiled 3D models—rather than by a learned or vision-language verifier—is sufficient to improve geometric accuracy. Formally, the rule is $$i^* = \arg\min_{i\in V} \frac{1}{|V|-1}\sum_{j\in V\setminus\{i\}} d(S_i,S_j),$$ with $d$ either the symmetric Chamfer distance between ICP-aligned, uniformly sampled surface point clouds (geometric consensus) or a 0-1 distance on Euler characteristic classes (topological consensus). On the shared watertight subsets of the EvoCAD initial pools, geometric consensus achieves Chamfer distance 0.0610, Hausdorff distance 0.176, and voxel IoU 0.703, versus 0.0627, 0.181, and 0.695 for the verifier, all significant at $p \le 0.022$; topological consensus shows no significant difference from the verifier on topology ($p = 0.82$ for Euler-characteristic error, $p = 1.00$ for topology-correct percentage). The paper interprets this as minimum Bayes risk decoding with the empirical distribution over samples, and notes that the oracle over the same pool remains clearly better.

Load-bearing premise

The rule rests on the premise that correct features recur across candidates while individual mistakes are random; if a pool shares systematic errors, consensus will select those errors, a failure the paper concedes in Section 5 and supports only with empirical evidence on CADPrompt, not a distributional argument.

Editorial extensions

If this is right

  • CAD verifiers used for candidate selection should be compared against consensus selection on the same candidate pool; on the EvoCAD pools, consensus matches or beats a vision-language verifier without any extra model calls.
  • Test-time scaling in CAD is not only about drawing more samples: selection quality saturates around $N \approx 9$, so larger pools mainly add coverage, while the oracle over the same pool stays well ahead.
  • The loss used for agreement determines which metric improves: geometric consensus with Chamfer distance helps geometry, and topological consensus with Euler characteristic helps topology, so the rule can be targeted to a desired property.
  • Because alignment normalizes away absolute dimensions, consensus does not currently exploit dimensional information even when prompts specify measurements; a direct corollary of the method's preprocessing.

Reading between the lines

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

  • A natural next test is to preserve outliers alongside the consensus candidate: returning the medoid plus the most distant candidate would recover high-quality outliers that agreement-based selection misses, a gap the paper attributes to the oracle.
  • The paper's loss-determines-metric result suggests other CAD-relevant properties—dimensions, manufacturability, or finite-element simulation outputs—could be plugged into the same rule; the authors mention this only as future work.
  • Whether consensus can form across sequential refinements of one program rather than parallel samples remains untested; if refinements share features, the same agreement rule might apply to iterative generation.
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 / 3 minor

Summary. The paper proposes a verifier-free selection method for text-to-CAD generation. Instead of using a vision-language verifier to choose among sampled CAD programs, the method compiles all candidates to 3D models and returns the candidate with greatest agreement with the rest of the pool: a geometric medoid under Chamfer distance, or a majority vote over Euler-characteristic classes. The method is framed as minimum Bayes risk decoding, is training-free, and is evaluated on the CADPrompt benchmark. On EvoCAD's initial candidate pools, geometric consensus improves all three geometric metrics over the verifier (with reported p-values), while topological consensus matches the verifier on topology. Across four LLMs and two prompt variants, geometric consensus is reported to improve over random selection by 1--10% in Chamfer distance, with an oracle still clearly better. The paper also studies scaling with pool size and discusses limitations, including the risk of selecting systematically shared errors.

Significance. If the empirical claims hold, this is a useful, low-cost alternative to VLM-based verifiers for CAD candidate selection, relevant to test-time scaling in code generation and CAD. The controlled same-pool comparison in Table 1 is a genuine strength: the candidates are identical and only the selection rule varies, making the geometric improvements credible. The method has no fitted parameters, is grounded in established MBR decoding ideas, and is clearly presented as an extension of execution-based consensus to compiled 3D models. The main risk is that the central mechanism — that correct features are common and mistakes are random — is only validated in aggregate, not per prompt, and the paper's own Section 5 identifies the systematic-error failure mode. If the method is to be positioned as a general replacement for verifiers, that mechanism needs direct support.

major comments (4)
  1. [Section 4.4, Table 2] The claim that geometric consensus improves over random selection for every tested LLM and prompt variant rests on point estimates without confidence intervals or significance tests. The smallest improvement is 0.7% (Gemini 3 Flash with measurements: CD 0.0546 vs 0.0550), which is plausibly within sampling noise given n=125 prompts and 1000 random subsets. Please report paired significance tests (e.g., bootstrap or Wilcoxon) and confidence intervals for each row, or qualify the claim to reflect that the gains are not all statistically significant.
  2. [Section 3, Section 5] The method's justification in Section 3 is that 'correct features appear in many candidates, while individual mistakes are random.' Section 5 correctly concedes that if the same error appears in most candidates, that error becomes part of the consensus. Since all candidates come from the same LLM, the same few-shot examples, and the same prompt, correlated errors are a concrete possibility rather than a remote one. The paper reports only aggregate improvements and gives no per-prompt analysis of whether agreement scores correlate with true geometric error. Without such an analysis, the central claim that consensus provides a reliable verifier-free signal is not established beyond the specific aggregate numbers. I recommend adding a per-prompt correlation analysis between the consensus score and the candidate's error, or a breakdown of failure cases, to test the stated principle.
  3. [Section 4.4] The evaluation protocol for Table 2 is ambiguous. The text says 'evaluate selection at N=15 with 1000 random subsets per prompt,' but the mean candidate count ('Cand.') is below 15 for several models (e.g., Gemma 3 12B: 13.3). It is unclear how subsets of size 15 are constructed when fewer than 15 candidates compile, whether prompts without enough candidates are excluded (and how many), or whether sampling is with replacement. Please specify the subset construction and report the number of prompts actually used per row.
  4. [Section 4.2, Abstract] The abstract and introduction state that consensus selection is evaluated 'on the exact candidate pools' of EvoCAD. In fact, the controlled comparison restricts all selection methods to the watertight subset of each initial pool (Section 4.2, Protocol). This is a reasonable choice for comparing topological consensus, but it changes the verifier's selection space relative to its original protocol. The claim should be qualified as applying to watertight subsets, and the verifier's performance on the full pool could be reported as a reference to assess the effect of this restriction.
minor comments (3)
  1. [Section 4.1] The voxel IoU metric is described as 'sample 10,000 points from each aligned mesh and convert them into a solid voxel grid with voxel size 0.1.' Please clarify how the points are converted to occupancy (e.g., whether each voxel is occupied if it contains at least one point) and whether the voxel grid origin is aligned across models.
  2. [Section 4.2] The significance tests are reported only as p-values without specifying the statistical test used (e.g., paired bootstrap, permutation test, Wilcoxon). Please state the test and whether any multiple-comparison correction was applied when testing several metrics.
  3. [Figure 2] The shaded bands are described as ±1 standard deviation over random subset draws, which is informative. The right panel's sample size is given only as '180 with watertight ground truth'; please also state the number of prompts for the left panel explicitly in the caption (the text says 182 for CD, but the caption should match).

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: consensus selection is an unfitted MBR medoid rule, and its 'agreement tracks correctness' premise is an empirical assumption, not an equation-level equivalence.

full rationale

The paper's selection rule, Equation (1), is a minimum Bayes risk decoder with a uniform empirical distribution, which the authors explicitly attribute to prior work (Kumar and Byrne, 2004; Eikema and Aziz, 2022). No fitted parameters are involved, so no fitted input is renamed as a prediction. Using Chamfer distance both as the selection loss and as one evaluation metric is an alignment of objectives, not circularity: the selector minimizes average distance to the candidate pool, while evaluation measures distance to ground truth; these are different targets, and the paper transparently notes that each consensus variant performs best on the metric aligned with its loss. The load-bearing assumption that 'correct features appear in many candidates, while individual mistakes are random' is an empirical claim about the candidate distribution, not an identity or construction; the paper itself flags the failure mode when systematic errors form the majority (Section 5). All comparisons use fixed, externally generated candidate pools from Preintner et al. and Alrashedy et al., and the baselines are other authors' systems. There is no self-citation chain, no imported uniqueness theorem, and no ansatz smuggled in via citation. The skeptic's concern about correlated errors is a correctness or validation risk, not circularity, and does not affect the circularity score.

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

The method adds no learned parameters and no invented entities. Its central premise is the consensus principle, which the paper states explicitly but does not prove. Reproducibility is limited by unspecified geometric preprocessing details, and the evaluation depends on several standard hyperparameters that are not derived from the data. The empirical nature of the claim means the ledger is light; the main burden is the unverified distributional assumption about candidate pools.

free parameters (4)
  • N (candidates sampled per prompt in evaluation) = 30 generated, 15 used for selection
    Evaluation hyperparameter chosen by the authors; the scaling study shows gains saturate near N=9, so N=15 is reasonable but not derived from the theory.
  • Point cloud sample count for Chamfer distance
    The method samples a point cloud from each surface but the text never states how many points are used; this affects the Chamfer estimate and is needed for reproduction.
  • ICP alignment parameters
    Each pair of models is normalized and aligned with ICP before distance computation, but the number of iterations, convergence criteria, and normalization scheme are unspecified.
  • Voxel size for IoU evaluation = 0.1
    Used to convert sampled points to a solid voxel grid; a standard evaluation choice that affects absolute IoU values but not the selection rule.
assumptions (3)
  • domain assumption Correct features appear in many candidates and individual mistakes are random.
    Stated at the start of Section 3 as the basis of consensus selection. If the candidate pool does not have this property, agreement no longer indicates correctness and the method selects the shared error.
  • domain assumption The CadQuery compiler faithfully converts programs into 3D models, and the mesh conversion preserves the properties used for agreement.
    The paper relies on the compiled model as the object of comparison; any compiler or meshing distortion would change the consensus signal. This is standard practice in the cited CAD generation literature.
  • standard math Sampling-based MBR with a uniform distribution over the pool approximates minimum expected loss (Eikema and Aziz, 2022).
    The decision rule in Equation (1) is justified as MBR; the approximation quality is inherited from prior work, not analyzed here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Test-Time Scaling for CAD Generation via Verifier-Free Consensus Selection." pith.science (2026). https://pith.science/paper/D4CLBRT7

@misc{pith2026260809706,
  author       = {Pith},
  title        = {Pith review of: Test-Time Scaling for CAD Generation via Verifier-Free Consensus Selection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D4CLBRT7}},
  note         = {Machine review of arXiv:2608.09706}
}
abstract

Large language models can write parametric CAD programs from a natural-language description (text-to-CAD generation), but a single sample is often wrong. Increasing test-time compute by sampling multiple candidates only helps if a good candidate can be identified, yet no ground-truth model is available at generation time. Existing systems often require a separate verifier, such as a vision-language judge, to select among candidates. We investigate whether the candidate pool itself provides enough signal for effective selection and a verifier-free alternative. We introduce 3D CAD consensus selection, hereafter consensus selection: sample $N$ parametric CAD programs, compile them to 3D models, and return the candidate that agrees most with the rest of the pool. The method is training-free and compatible with existing CAD agents. We investigate geometric and topological notions of agreement, each of which improves its corresponding evaluation metric. On the exact candidate pools of a state-of-the-art CAD generation method, geometric consensus improves all three geometric metrics over the method's verifier, while topological consensus matches it on topology. Across every tested LLM and prompt variant, geometric consensus also improves geometric accuracy over random selection from the same pool, reducing Chamfer distance by $1-10\%$.

Figures

Figures reproduced from arXiv: 2608.09706 by the authors.

Figure 1
Figure 1. Consensus selection for 3D CAD generation. We compile N sampled parametric CAD programs into 3D models and select one based on agreement in their 3D CAD model properties. Problem setting. The task is to generate a parametric CAD program c from a natural-language prompt p; a compiler ϕ converts the program into a 3D model ϕ(c) [Preintner et al., 2025] facilitating a direct comparison of properties of 3D models. The g… view at source ↗
Figure 2
Figure 2. Scaling with the number of sampled candidates [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 7 canonical work pages

  1. [1]

    o nig, Adrian and B \

    Preintner, Tobias and Yuan, Weixuan and K \" o nig, Adrian and B \" a ck, Thomas and Raponi, Elena and van Stein, Niki. EvoCAD : Evolutionary CAD Code Generation with Vision Language Models. 2025 IEEE 37th International Conference on Tools with Artificial Intelligence (ICTAI). doi:10.1109/ictai66417.2025.00075

    arXiv 2025

  2. [2]

    and Langwasser, Megan and Xu, Wei and Gombolay, Matthew C

    Alrashedy, Kamel and Tambwekar, Pradyumna and Zaidi, Z. and Langwasser, Megan and Xu, Wei and Gombolay, Matthew C. Generating CAD Code with Vision-Language Models for 3D Designs. International Conference on Learning Representations. doi:10.48550/arxiv.2410.05340

    arXiv doi:10.48550/arxiv.2410.05340

  3. [3]

    3D-Premise : Can Large Language Models Generate 3D Shapes with Sharp Features and Parametric Control?

    Yuan, Zeqing and Lan, Haoxuan and Zou, Qiang and Zhao, Junbo. 3D-Premise : Can Large Language Models Generate 3D Shapes with Sharp Features and Parametric Control?. arXiv preprint arXiv:2401.06437. doi:10.48550/arxiv.2401.06437. 2401.06437

    arXiv doi:10.48550/arxiv.2401.06437

  4. [4]

    ArtiCAD : Articulated CAD Assembly Design via Multi-Agent Code Generation

    Shui, Yuan and Guan, Yandong and Zhang, Zhanwei and Hu, Juncheng and Zhang, Jing and Xu, Dong and Yu, Qian. ArtiCAD : Articulated CAD Assembly Design via Multi-Agent Code Generation. arXiv preprint arXiv:2604.10992. doi:10.48550/arxiv.2604.10992. 2604.10992

    arXiv doi:10.48550/arxiv.2604.10992

  5. [5]

    MEDA : A Multi-Agent System for Parametric CAD Model Creation

    Panta, Nirmal and Kafley, Saugat and Acharya, Rujal and Parajuli, Sashank and Parajuli, Dikshya and Panta, Prince and Belbase, Saroj and Pant, Sudikshya and Regmi, Amit and Tanaka, Akio and McComb, Christopher. MEDA : A Multi-Agent System for Parametric CAD Model Creation. Volume 3B: 51st Design Automation Conference (DAC). doi:10.1115/detc2025-163946

    doi:10.1115/detc2025-163946

  6. [6]

    CADSmith : Multi-Agent CAD Generation with Programmatic Geometric Validation

    Barkley, Jesse and Loghmani, Rumi and Farimani, Amir Barati. CADSmith : Multi-Agent CAD Generation with Programmatic Geometric Validation. arXiv preprint arXiv:2603.26512. doi:10.48550/arxiv.2603.26512. 2603.26512

    doi:10.48550/arxiv.2603.26512

  7. [7]

    MUSE : Benchmarking Manufacturable, Functional, and Assemblable Text-to- CAD Generation

    Dong, Xiaoyu and Li, Zhi and Wu, Xiao-Ming. MUSE : Benchmarking Manufacturable, Functional, and Assemblable Text-to- CAD Generation. arXiv preprint arXiv:2605.28579. doi:10.48550/arxiv.2605.28579. 2605.28579

    arXiv doi:10.48550/arxiv.2605.28579

  8. [8]

    Query2CAD : Generating CAD Models Using Natural Language Queries

    Badagabettu, Akshay and Yarlagadda, Sai Sravan and Farimani, Amir Barati. Query2CAD : Generating CAD Models Using Natural Language Queries. arXiv preprint arXiv:2406.00144. doi:10.48550/arxiv.2406.00144. 2406.00144

    arXiv doi:10.48550/arxiv.2406.00144

Show all 42 references
  1. [9]

    Seek-CAD : A Self-Refined Generative Modeling for 3D Parametric CAD Using Local Inference via DeepSeek

    Li, Xueyang and Li, Jiahao and Song, Yu and Lou, Yunzhong and Zhou, Xiangdong. Seek-CAD : A Self-Refined Generative Modeling for 3D Parametric CAD Using Local Inference via DeepSeek. International Conference on Learning Representations (ICLR). doi:10.48550/arxiv.2505.17702

    doi

  2. [10]

    and Lin, Dahua and Wetzstein, Gordon

    Wu, Tong and Yang, Guandao and Li, Zhibing and Zhang, Kai and Liu, Ziwei and Guibas, Leonidas J. and Lin, Dahua and Wetzstein, Gordon. GPT-4V (ision) is a Human-Aligned Evaluator for Text-to- 3D Generation. Computer Vision and Pattern Recognition. doi:10.1109/cvpr52733.2024.02098

  3. [12]

    and Company, Pedro

    Zhou, Jiwei and Camba, Jorge D. and Company, Pedro. CADialogue : A Multimodal LLM -Powered Conversational Assistant for Intuitive Parametric CAD Modeling. Computer-Aided Design. doi:10.1016/j.cad.2025.104006

    2025

  4. [13]

    Text-to- CAD Generation Through Infusing Visual Feedback in Large Language Models

    Wang, Ruiyu and Yuan, Yu and Sun, Shizhao and Bian, Jiang. Text-to- CAD Generation Through Infusing Visual Feedback in Large Language Models. International Conference on Machine Learning. doi:10.48550/arxiv.2501.19054

    doi arXiv

  5. [14]

    and Heyrani Nobari, Amin and Xu, Kai and Srivastava, Akash and Ahmed, Faez

    Giannone, Giorgio and Doris, Anna C. and Heyrani Nobari, Amin and Xu, Kai and Srivastava, Akash and Ahmed, Faez. GIFT : Bootstrapping Image-to- CAD Program Synthesis via Geometric Feedback. International Conference on Machine Learning. doi:10.48550/arxiv.2603.27448. 2603.27448

    doi

  6. [15]

    CAD-Judge : Toward Efficient Morphological Grading and Verification for Text-to- CAD Generation

    Zhou, Zheyuan and Han, Jiayi and Du, Liang and Fang, Naiyu and Qiu, Lemiao and Zhang, Shuyou. CAD-Judge : Toward Efficient Morphological Grading and Verification for Text-to- CAD Generation. ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Proce...

    2026

  7. [16]

    2026 , howpublished =

    2026

  8. [17]

    Self-Improving CAD Generation Agents with Finite Element Analysis as Feedback

    Son, Guijin and Park, Jehyun and Park, Seyeon and Ahn, Sunghee and Yu, Youngjae. Self-Improving CAD Generation Agents with Finite Element Analysis as Feedback. arXiv preprint arXiv:2605.17448. doi:10.48550/arxiv.2605.17448. 2605.17448

    doi arXiv

  9. [18]

    Physics-in-the-Loop: A Hybrid Agentic Architecture for Validated CAD Engineering Design

    Berger, Elias and Usama, Muhammad and Mehlst \" a ubl, Jan and Saske, Bernhard and Paetzold-Byhain, Kristin. Physics-in-the-Loop: A Hybrid Agentic Architecture for Validated CAD Engineering Design. arXiv preprint arXiv:2605.19717. doi:10.48550/arxiv.2605.19717. 2605.19717

    doi arXiv

  10. [19]

    Embodied CAD : Solver-Grounded LLM Agents for Parametric B -Rep Assembly Modeling

    Liu, Fumin and Zhou, Haoyu and Hao, Fei and Yang, Lin. Embodied CAD : Solver-Grounded LLM Agents for Parametric B -Rep Assembly Modeling. arXiv preprint arXiv:2606.31252. doi:10.48550/arxiv.2606.31252. 2606.31252

    doi arXiv

  11. [20]

    and Jiang, Liwei and Fisher, Jillian R

    West, Peter and Lu, Ximing and Dziri, Nouha and Brahman, Faeze and Li, Linjie and Hwang, Jena D. and Jiang, Liwei and Fisher, Jillian R. and Ravichander, Abhilasha and Chandu, Khyathi Raghavi and Newman, Benjamin and Koh, Pang Wei and Ettinger, Allyson and Choi, Yejin. The Gen...

    doi arXiv

  12. [21]

    and Zhang, Hao and Gonzalez, Joseph E

    Zheng, Lianmin and Chiang, Wei-Lin and Sheng, Ying and Zhuang, Siyuan and Wu, Zhanghao and Zhuang, Yonghao and Lin, Zi and Li, Zhuohan and Li, Dacheng and Xing, E. and Zhang, Hao and Gonzalez, Joseph E. and Stoica, Ion. Judging LLM -as-a-Judge with MT -Bench and Chatbot Arena....

    doi

  13. [22]

    and Feng, Shi

    Panickssery, Arjun and Bowman, Samuel R. and Feng, Shi. LLM Evaluators Recognize and Favor Their Own Generations. arXiv preprint arXiv:2404.13076. doi:10.48550/arxiv.2404.13076

    doi arXiv

  14. [23]

    Scaling Laws for Reward Model Overoptimization

    Gao, Leo and Schulman, John and Hilton, Jacob. Scaling Laws for Reward Model Overoptimization. International Conference on Machine Learning. doi:10.48550/arxiv.2210.10760

    doi arXiv

  15. [24]

    and Narang, Sharan and Chowdhery, Aakanksha and Zhou, Denny

    Wang, Xuezhi and Wei, Jason and Schuurmans, Dale and Le, Quoc and Chi, Ed H. and Narang, Sharan and Chowdhery, Aakanksha and Zhou, Denny. Self-Consistency Improves Chain of Thought Reasoning in Language Models. International Conference on Learning Representations. doi:10.48550...

    doi arXiv

  16. [25]

    It's MBR All the Way Down: Modern Generation Techniques Through the Lens of Minimum B ayes Risk

    Bertsch, Amanda and Xie, Alex and Neubig, Graham and Gormley, Matthew. It's MBR All the Way Down: Modern Generation Techniques Through the Lens of Minimum B ayes Risk. Proceedings of the Big Picture Workshop. doi:10.18653/v1/2023.bigpicture-1.9

    doi

  17. [26]

    DeepCAD : A Deep Generative Network for Computer-Aided Design Models

    Wu, Rundi and Xiao, Chang and Zheng, Changxi. DeepCAD : A Deep Generative Network for Computer-Aided Design Models. 2021 IEEE/CVF International Conference on Computer Vision (ICCV). doi:10.1109/iccv48922.2021.00670

    2021

  18. [27]

    FlexCAD : Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models

    Zhang, Zhanwei and Sun, Shizhao and Wang, Wenxiao and Cai, Deng and Bian, Jiang. FlexCAD : Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models. International Conference on Learning Representations. doi:10.48550/arxiv.2411.05823

    doi arXiv

  19. [28]

    CAD-Llama : Leveraging Large Language Models for Computer-Aided Design Parametric 3 D Model Generation

    Li, Jiahao and Ma, Weijian and Li, Xueyang and Lou, Yunzhong and Zhou, Guichun and Zhou, Xiangdong. CAD-Llama : Leveraging Large Language Models for Computer-Aided Design Parametric 3 D Model Generation. 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR...

    2025

  20. [29]

    cadrille: Multi-modal CAD Reconstruction with Reinforcement Learning

    Kolodiazhnyi, Maksim and Tarasov, Denis and Zhemchuzhnikov, Dmitrii and Nikulin, Alexander and Zisman, Ilya and Vorontsova, Anna and Konushin, Anton and Kurenkov, Vladislav and Rukhovich, Danila. cadrille: Multi-modal CAD Reconstruction with Reinforcement Learning. arXiv prepr...

    doi

  21. [30]

    and Cao, Yuan and Narasimhan, Karthik

    Yao, Shunyu and Yu, Dian and Zhao, Jeffrey and Shafran, Izhak and Griffiths, T. and Cao, Yuan and Narasimhan, Karthik. Tree of Thoughts: Deliberate Problem Solving with Large Language Models. Neural Information Processing Systems. doi:10.48550/arxiv.2305.10601

    doi arXiv

  22. [31]

    Minimum B ayes-Risk Decoding for Statistical Machine Translation

    Kumar, Shankar and Byrne, W. Minimum B ayes-Risk Decoding for Statistical Machine Translation. North American Chapter of the Association for Computational Linguistics

  23. [32]

    Sampling-Based Approximations to Minimum B ayes Risk Decoding for Neural Machine Translation

    Eikema, Bryan and Aziz, Wilker. Sampling-Based Approximations to Minimum B ayes Risk Decoding for Neural Machine Translation. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. doi:10.18653/v1/2022.emnlp-main.754

    2022 doi

  24. [33]

    Natural Language to Code Translation with Execution

    Shi, Freda and Fried, Daniel and Ghazvininejad, Marjan and Zettlemoyer, Luke and Wang, Sida I. Natural Language to Code Translation with Execution. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. doi:10.18653/v1/2022.emnlp-main.231

    2022 doi

  25. [34]

    Universal Self-Consistency for Large Language Model Generation

    Chen, Xinyun and Aksitov, Renat and Alon, Uri and Ren, Jie and Xiao, Kefan and Yin, Pengcheng and Prakash, Sushant and Sutton, Charles and Wang, Xuezhi and Zhou, Denny. Universal Self-Consistency for Large Language Model Generation. arXiv preprint arXiv:2311.17311. doi:10.4855...

    doi arXiv

  26. [35]

    and R\' e , Christopher and Mirhoseini, Azalia

    Brown, Bradley and Juravsky, Jordan and Ehrlich, Ryan and Clark, Ronald and Le, Quoc V. and R\' e , Christopher and Mirhoseini, Azalia. Large Language Monkeys: Scaling Inference Compute with Repeated Sampling. arXiv preprint arXiv:2407.21787. doi:10.48550/arxiv.2407.21787. 2407.21787

    doi arXiv

  27. [36]

    and Dupont, Elona and Mallis, Dimitrios and Cherenkova, K

    Rukhovich, D. and Dupont, Elona and Mallis, Dimitrios and Cherenkova, K. and Kacem, Anis and Aouada, D. CAD-Recode : Reverse Engineering CAD Code from Point Clouds. IEEE International Conference on Computer Vision. doi:10.1109/iccv51701.2025.00914

  28. [37]

    Competition-Level Code Generation with AlphaCode

    Li, Yujia and Choi, David and Chung, Junyoung and Kushman, Nate and Schrittwieser, Julian and Leblond, R\' e mi and Eccles, Tom and Keeling, James and Gimeno, Felix and Dal Lago, Agustin and Hubert, Thomas and Choy, Peter and de Masson d'Autume, Cyprien and Babuschkin, Igor an...

  29. [38]

    and Stricker, Didier and Ali, Sk Aziz and Afzal, Muhammad Zeshan

    Khan, Mohammad Sadil and Sinha, Sankalp and Sheikh, T. and Stricker, Didier and Ali, Sk Aziz and Afzal, Muhammad Zeshan. Text2CAD : Generating Sequential CAD Designs from Beginner-to-Expert Level Text Prompts. Neural Information Processing Systems. doi:10.52202/079017-0242

    doi

  30. [39]

    Text-to- CadQuery : A New Paradigm for CAD Generation with Scalable Large Model Capabilities

    Xie, Haoyang and Ju, Feng. Text-to- CadQuery : A New Paradigm for CAD Generation with Scalable Large Model Capabilities. arXiv preprint arXiv:2505.06507. doi:10.48550/arxiv.2505.06507. 2505.06507

    doi arXiv

  31. [40]

    CAD-Coder : Text-to- CAD Generation with Chain-of-Thought and Geometric Reward

    Guan, Yandong and Wang, Xilin and Xing, Ximing and Zhang, Jing and Xu, Dong and Yu, Qian. CAD-Coder : Text-to- CAD Generation with Chain-of-Thought and Geometric Reward. Neural Information Processing Systems. doi:10.48550/arxiv.2505.19713. 2505.19713

    doi arXiv

  32. [41]

    2026 , publisher=

    Doris, Anna C and Alam, Md Ferdous and Heyrani Nobari, Amin and Ahmed, Faez , journal=. 2026 , publisher=

    2026

  33. [42]

    Towards High-Fidelity CAD Generation via LLM -Driven Program Generation and Text-Based B -Rep Primitive Grounding

    Li, Jiahao and Zhang, Qingwang and Chen, Qiuyu and Qiu, Guozhan and Lou, Yunzhong and Zhou, Xiangdong. Towards High-Fidelity CAD Generation via LLM -Driven Program Generation and Text-Based B -Rep Primitive Grounding. International Conference on Machine Learning (ICML). doi:10...

    doi arXiv

  34. [43]

    Text2CAD-Bench : A Benchmark for LLM -based Text-to-Parametric CAD Generation

    Wang, Liang and Meng, Heng and Xiang, Zekai and Liu, Jin and Zhou, Pingyi and Chen, Litao and Tang, Yongqiang. Text2CAD-Bench : A Benchmark for LLM -based Text-to-Parametric CAD Generation. arXiv preprint arXiv:2605.18430. doi:10.48550/arxiv.2605.18430. 2605.18430

    doi arXiv

Pith tools

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