pith. machine review for the scientific record. sign in

arxiv: 2604.24464 · v1 · submitted 2026-04-27 · 💻 cs.DC

Recognition: unknown

Incisor: Ex Ante Cloud Instance Selection for HPC Jobs

David A. B. Hyde, Michael A. Laurenzano, Shihan Cheng

Authors on Pith no claims yet

Pith reviewed 2026-05-07 17:54 UTC · model grok-4.3

classification 💻 cs.DC
keywords incisorinstancecloudantehardwareselectionsubmissionanalysis
0
0 comments X

The pith

Incisor uses program analysis and frontier LLMs to select working AWS EC2 instances ex ante for 100% of first-time HPC runs of C/C++/Fortran and Python codes, cutting runtime 54% and costs 44% versus an expert-constrained SkyPilot baseline.

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

The problem is choosing the right cloud computer before you run a job. Today this is done by hand: users look at their code, guess what hardware it needs, and pick from hundreds of instance types. Incisor tries to automate that guess. It runs standard program analysis on the source code and then asks a large language model to reason about memory, CPU, and other requirements. The system then picks an instance type that should work. In the reported tests it succeeded on every first-time run of compiled and Python HPC applications. Compared with a baseline that already uses expert rules plus an existing cloud tool, the chosen instances ran the jobs faster and cheaper.

Core claim

Using submission artifacts alone, Incisor atop frontier coding LLMs selects working AWS EC2 instances ex ante for 100% of first-time runs of source-compiled (C, C++, Fortran) and Python applications. Against a strong baseline combining expert-derived constraints with SkyPilot's instance selection, Incisor cuts job runtime by 54% and instance costs by 44%.

Load-bearing premise

That LLM-guided reasoning over program-analysis outputs can reliably infer hardware constraints and produce instance selections that are both correct and cost-effective without any runtime profiling or additional user-provided hints.

Figures

Figures reproduced from arXiv: 2604.24464 by David A. B. Hyde, Michael A. Laurenzano, Shihan Cheng.

Figure 1
Figure 1. Figure 1: The number of available instance types from the major cloud providers view at source ↗
Figure 2
Figure 2. Figure 2: Memory capacity distribution across 794 x86_64 Linux AWS instance types. Instances below the 18 GB threshold (red x pattern) for CG class D are expected to fail with an out-of-memory error. ROCm), and offload-related code reveals no evidence of paral￾lel I/O or accelerator usage, ruling out benefits from premium storage and GPU-enabled instances. Instance Selection. Taken together, this evidence defines a … view at source ↗
Figure 3
Figure 3. Figure 3: Incisor agent architecture operation for hardware constraint estimation. view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the Incisor system’s job handling steps. view at source ↗
Figure 5
Figure 5. Figure 5: Job success rates using Incisor agent selections and SkyPilot baseline view at source ↗
Figure 6
Figure 6. Figure 6: Incisor constraint estimates across input configurations. (a) Fraction of job submissions with over- or under-provisioning risk, showing that tool access view at source ↗
Figure 7
Figure 7. Figure 7: Instance costs for the top 5 Incisor-chosen instances compared to the view at source ↗
Figure 8
Figure 8. Figure 8: Runtime for the top 5 Incisor-selected instances normalized to the view at source ↗
read the original abstract

We present Incisor, a cloud HPC job submission system for the ex ante instance selection problem: choosing suitable hardware in the challenging but common setting where only the executable, inputs, and invocation commands are available at submission time. In practice, this task is manual and expertise-intensive, requiring users to combine incomplete knowledge of rapidly evolving cloud offerings with workload-specific intuition, static analysis, and systems reasoning to infer hardware constraints and select an instance type for each job. Incisor automates this process by pairing widely available program analysis tools with LLM-guided reasoning to infer hardware requirements and choose cloud instances. Using submission artifacts alone, Incisor atop frontier coding LLMs selects working AWS EC2 instances ex ante for 100% of first-time runs of source-compiled (C, C++, Fortran) and Python applications. Against a strong baseline combining expert-derived constraints with SkyPilot's instance selection, Incisor cuts job runtime by 54% and instance costs by 44%.

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.

Referee Report

2 major / 1 minor

Summary. Incisor is a system for ex ante cloud instance selection in HPC that pairs program analysis tools with LLM-guided reasoning to infer hardware requirements from only the executable, inputs, and invocation commands. The paper claims that Incisor atop frontier coding LLMs achieves 100% success in selecting working AWS EC2 instances for first-time runs of C/C++/Fortran and Python applications, while cutting runtime by 54% and costs by 44% versus a baseline of expert-derived constraints plus SkyPilot.

Significance. If the empirical results hold under rigorous validation, the work could meaningfully lower the expertise barrier for cloud HPC by automating instance selection. The combination of static analysis with LLM reasoning is a timely approach for systems that must operate without profiling or user hints. The reported gains indicate practical efficiency potential, but the current presentation leaves the robustness of the 100% success rate and percentage improvements difficult to assess.

major comments (2)
  1. [Abstract and Evaluation] Abstract and Evaluation section: the headline claims of 100% success, 54% runtime reduction, and 44% cost reduction are presented without any description of the number of jobs, workload diversity, statistical significance testing, controls for instance availability, or LLM variability. This information is load-bearing for the central empirical contribution and must be supplied to allow readers to evaluate generalizability.
  2. [System Design] System design and assumption discussion: the approach rests on LLM reasoning over static program-analysis outputs being sufficient to deduce dynamic constraints such as peak memory, core scaling, or communication volume. The manuscript does not provide evidence or ablation that this inference step succeeds for input-dependent behaviors, which directly supports the 100% success claim and the reported gains over the SkyPilot baseline.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by a single sentence stating the scale of the evaluation (e.g., number of distinct applications or total job runs) to give immediate context for the quantitative results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback on our manuscript. We have revised the paper to address the concerns about the presentation of empirical claims and the discussion of key assumptions in the system design. Our point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract and Evaluation] Abstract and Evaluation section: the headline claims of 100% success, 54% runtime reduction, and 44% cost reduction are presented without any description of the number of jobs, workload diversity, statistical significance testing, controls for instance availability, or LLM variability. This information is load-bearing for the central empirical contribution and must be supplied to allow readers to evaluate generalizability.

    Authors: We agree that these details are necessary for readers to properly evaluate the generalizability of our results. The original Evaluation section describes the individual applications and reports aggregate metrics but does not consolidate the requested information or include statistical tests. In the revised manuscript we have added a new 'Experimental Setup' subsection that supplies the number of jobs evaluated, workload diversity (including application domains and input variations), statistical significance testing for the reported runtime and cost improvements, controls for instance availability on AWS, and steps taken to account for LLM variability. We have also updated the abstract to briefly reference the scale of the evaluation. These changes directly respond to the comment. revision: yes

  2. Referee: [System Design] System design and assumption discussion: the approach rests on LLM reasoning over static program-analysis outputs being sufficient to deduce dynamic constraints such as peak memory, core scaling, or communication volume. The manuscript does not provide evidence or ablation that this inference step succeeds for input-dependent behaviors, which directly supports the 100% success claim and the reported gains over the SkyPilot baseline.

    Authors: We appreciate the referee highlighting this core assumption. Section 4 of the manuscript describes how the LLM is prompted with static-analysis outputs to reason about likely dynamic behaviors, including input-dependent ones, by leveraging common HPC code patterns. The 100% success rate is measured on workloads that include input variation, providing indirect support. We acknowledge the absence of a dedicated ablation isolating the LLM's contribution to input-dependent inference. In the revision we have expanded the System Design section with additional discussion and qualitative examples from our case studies showing how the LLM infers scaling from code structure. We note that a quantitative ablation would strengthen the claim further and list this as future work, but the current evidence from end-to-end results demonstrates the practical utility of the combined static-analysis plus LLM approach. revision: partial

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the untested premise that current frontier LLMs can perform reliable hardware-requirement inference from static analysis output alone. No free parameters or invented physical entities are introduced.

axioms (1)
  • domain assumption Frontier coding LLMs can accurately translate program-analysis facts into hardware constraints and instance-type recommendations without runtime data.
    This assumption is required for the 100% success claim and is not independently verified in the abstract.

pith-pipeline@v0.9.0 · 5467 in / 1335 out tokens · 68516 ms · 2026-05-07T17:54:45.510027+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 54 canonical work pages · 5 internal anchors

  1. [1]

    Predictive performance and scalabil- ity modeling of a large-scale application,

    D. J. Kerbyson, H. J. Alme, A. Hoisie, F. Petrini, H. J. Wasserman, and M. Gittings, “Predictive performance and scalabil- ity modeling of a large-scale application,” inProceedings of the 2001 ACM/IEEE conference on Supercomputing, 2001, pp. 37–37, https://doi.org/10.1145/582034.582071

    work page doi:10.1145/582034.582071 2001

  2. [2]

    A framework for performance modeling and prediction,

    A. Snavely, L. Carrington, N. Wolter, J. Labarta, R. Badia, and A. Purkayastha, “A framework for performance modeling and prediction,” inSC’02: Proceedings of the 2002 ACM/IEEE Conference on Supercomputing. IEEE, 2002, pp. 21–21, https://doi.org/10.1109/sc.2002.10004

    work page doi:10.1109/sc.2002.10004 2002

  3. [3]

    An ap- proach to performance prediction for parallel applications,

    E. Ipek, B. R. De Supinski, M. Schulz, and S. A. McKee, “An ap- proach to performance prediction for parallel applications,” inEuropean Conference on Parallel Processing. Springer, 2005, pp. 196–205, https://doi.org/10.1007/11549468 24

    work page doi:10.1007/11549468 2005

  4. [4]

    Roofline: an in- sightful visual performance model for multicore architectures,

    S. Williams, A. Waterman, and D. Patterson, “Roofline: an in- sightful visual performance model for multicore architectures,”Com- munications of the ACM, vol. 52, no. 4, pp. 65–76, 2009, https://doi.org/10.1145/1498765.1498785

    work page doi:10.1145/1498765.1498785 2009

  5. [5]

    Using automated performance modeling to find scalability bugs in complex codes,

    A. Calotoiu, T. Hoefler, M. Poke, and F. Wolf, “Using automated performance modeling to find scalability bugs in complex codes,” inProceedings of the International Conference on High Perfor- mance Computing, Networking, Storage and Analysis, 2013, pp. 1–12, https://doi.org/10.1145/2503210.2503277

    work page doi:10.1145/2503210.2503277 2013

  6. [6]

    Palm: Easing the burden of an- alytical performance modeling,

    N. R. Tallent and A. Hoisie, “Palm: Easing the burden of an- alytical performance modeling,” inProceedings of the 28th ACM international conference on Supercomputing, 2014, pp. 221–230, https://doi.org/10.1145/2597652.2597683

    work page doi:10.1145/2597652.2597683 2014

  7. [7]

    Ernest: Efficient performance prediction for Large-Scale advanced analytics,

    S. Venkataraman, Z. Yang, M. Franklin, B. Recht, and I. Stoica, “Ernest: Efficient performance prediction for Large-Scale advanced analytics,” in13th USENIX Symposium on Networked Systems Design and Im- plementation (NSDI 16). USENIX Association, 2016, pp. 363–378, https://dl.acm.org/doi/10.5555/2930611.2930635

    work page doi:10.5555/2930611.2930635 2016

  8. [8]

    Predicting workflow task execution time in the cloud using a two-stage machine learning approach,

    T.-P. Pham, J. J. Durillo, and T. Fahringer, “Predicting workflow task execution time in the cloud using a two-stage machine learning approach,”IEEE Transactions on Cloud Computing, vol. 8, no. 1, pp. 256–268, 2017, https://doi.org/10.1109/TCC.2017.2732344

    work page doi:10.1109/tcc.2017.2732344 2017

  9. [9]

    CherryPick: Adaptively unearthing the best cloud configura- tions for big data analytics,

    O. Alipourfard, H. H. Liu, J. Chen, S. Venkataraman, M. Yu, and M. Zhang, “CherryPick: Adaptively unearthing the best cloud configura- tions for big data analytics,” in14th USENIX Symposium on Networked Systems Design and Implementation (NSDI 17). USENIX Association, 2017, pp. 469–482, https://dl.acm.org/doi/10.5555/3154630.3154669

    work page doi:10.5555/3154630.3154669 2017

  10. [10]

    Towards col- laborative continuous benchmarking for hpc,

    O. Pearce, A. Scott, G. Becker, R. Haque, N. Hanford, S. Brink, D. Jacobsen, H. Poxon, J. Domke, and T. Gamblin, “Towards col- laborative continuous benchmarking for hpc,” inProceedings of the SC’23 Workshops of The International Conference on High Perfor- mance Computing, Network, Storage, and Analysis, 2023, pp. 627–635, https://doi.org/10.1145/3624062.3624135

    work page doi:10.1145/3624062.3624135 2023

  11. [11]

    Adviser: An intuitive multi-cloud platform for scientific and ml workflows,

    S. Cheng, M. A. Laurenzano, B. Strauch, T. A. Ellis, K. Wadhwani, and D. A. B. Hyde, “Adviser: An intuitive multi-cloud platform for scientific and ml workflows,”arXiv preprint arXiv:2603.20941, 2026, https://doi.org/10.48550/arXiv.2603.20941

    work page doi:10.48550/arxiv.2603.20941 2026

  12. [12]

    SkyPilot: An intercloud broker for sky computing,

    Z. Yang, Z. Wu, M. Luo, W.-L. Chiang, R. Bhardwaj, W. Kwon, S. Zhuang, F. S. Luan, G. Mittal, S. Shenkeret al., “SkyPilot: An intercloud broker for sky computing,” in20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23). USENIX Association, 2023, pp. 437–455, https://www.usenix.org/conference/nsdi23/presentation/yang-zongheng

    2023

  13. [13]

    The nas parallel bench- marks—summary and preliminary results,

    D. H. Bailey, E. Barszcz, J. T. Barton, D. S. Browning, R. L. Carter, L. Dagum, R. A. Fatoohi, P. O. Frederickson, T. A. Lasinski, R. S. Schreiberet al., “The nas parallel bench- marks—summary and preliminary results,” inProceedings of the 1991 ACM/IEEE Conference on Supercomputing, 1991, pp. 158–165, https://doi.org/10.1145/125826.125925

    work page doi:10.1145/125826.125925 1991

  14. [14]

    Assisting static analysis with large language models: A chatgpt experiment,

    H. Li, Y . Hao, Y . Zhai, and Z. Qian, “Assisting static analysis with large language models: A chatgpt experiment,” inProceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, 2023, pp. 2107–2111, https://doi.org/10.1145/3611643.3613078

    work page doi:10.1145/3611643.3613078 2023

  15. [15]

    The emergence of large language models in static anal- ysis: A first look through micro-benchmarks,

    A. P. S. Venkatesh, S. Sabu, A. M. Mir, S. Reis, and E. Bod- den, “The emergence of large language models in static anal- ysis: A first look through micro-benchmarks,” inProceedings of the 2024 IEEE/ACM First International Conference on AI Foundation Models and Software Engineering, 2024, pp. 35–39, https://doi.org/10.1145/3650105.3652288

    work page doi:10.1145/3650105.3652288 2024

  16. [16]

    Interleaving static analysis and llm prompting,

    P. J. Chapman, C. Rubio-Gonz ´alez, and A. V . Thakur, “Interleaving static analysis and llm prompting,” inProceedings of the 13th ACM SIGPLAN International Workshop on the State Of the Art in Program Analysis, 2024, pp. 9–17, https://doi.org/10.1145/3652588.3663317

    work page doi:10.1145/3652588.3663317 2024

  17. [17]

    Llmdfa: analyzing dataflow in code with large language models,

    C. Wang, W. Zhang, Z. Su, X. Xu, X. Xie, and X. Zhang, “Llmdfa: analyzing dataflow in code with large language models,”Advances in Neural Information Processing Systems, vol. 37, pp. 131 545–131 574, 2024, https://doi.org/10.52202/079017-4181

    work page doi:10.52202/079017-4181 2024

  18. [18]

    A contemporary sur- vey of large language model assisted program analysis,

    J. Wang, T. Ni, W.-B. Lee, and Q. Zhao, “A contemporary sur- vey of large language model assisted program analysis,”Transac- tions on Artificial Intelligence, vol. 1, no. 1, pp. 105–129, 2025, https://doi.org/10.53941/tai.2025.100006

    work page doi:10.53941/tai.2025.100006 2025

  19. [19]

    Can large language models predict parallel code performance?

    G. Bolet, G. Georgakoudis, H. Menon, K. Parasyris, N. Hasabnis, H. Estes, K. Cameron, and G. Oren, “Can large language models predict parallel code performance?” inProceedings of the 34th International Symposium on High-Performance Parallel and Distributed Computing, 2025, pp. 1–6, https://doi.org/10.1145/3731545.3743645

    work page doi:10.1145/3731545.3743645 2025

  20. [20]

    Hpc-coder: Modeling parallel programs using large language mod- els,

    D. Nichols, A. Marathe, H. Menon, T. Gamblin, and A. Bhatele, “Hpc-coder: Modeling parallel programs using large language mod- els,” inISC High Performance 2024 Research Paper Proceedings (39th International Conference). Prometeus GmbH, 2024, pp. 1–12, https://doi.org/10.23919/isc.2024.10528929

    work page doi:10.23919/isc.2024.10528929 2024

  21. [21]

    Do large language models understand performance optimization?

    B. Cui, T. Ramesh, O. Hernandez, and K. Zhou, “Do large lan- guage models understand performance optimization?”arXiv preprint arXiv:2503.13772, 2025, https://doi.org/10.48550/arXiv.2503.13772

    work page doi:10.48550/arxiv.2503.13772 2025

  22. [22]

    Llmperf: Gpu performance modeling meets large language models,

    M.-K. Nguyen-Nhat, H. D. N. Do, H. T. Le, and T. T. Dao, “Llmperf: Gpu performance modeling meets large language models,” in2024 32nd International Conference on Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS). IEEE, 2024, pp. 1–8, https://doi.org/10.1109/mascots64422.2024.10786558

    work page doi:10.1109/mascots64422.2024.10786558 2024

  23. [23]

    Llm-based cost-aware task scheduling for cloud computing sys- tems,

    H. Pei, Y . Gu, Y . Sun, Q. Wang, C. Liu, X. Chen, and L. Cheng, “Llm-based cost-aware task scheduling for cloud computing sys- tems,”Journal of Cloud Computing, vol. 14, no. 1, p. 81, 2025, https://doi.org/10.1186/s13677-025-00822-0

    work page doi:10.1186/s13677-025-00822-0 2025

  24. [24]

    Exploring cloud instance options for optimal performance-cost efficiency,

    W. Maas and A. Lorenzon, “Exploring cloud instance options for optimal performance-cost efficiency,”Cluster Computing, vol. 28, no. 15, p. 984, 2025, https://doi.org/10.1007/s10586-025-05702-5

    work page doi:10.1007/s10586-025-05702-5 2025

  25. [25]

    Investigating cloud instances to achieve optimal trade-offs between performance-cost efficiency,

    W. Maas, F. Rossi, M. Caggiani, P. Navaux, and A. Lorenzon, “Investigating cloud instances to achieve optimal trade-offs between performance-cost efficiency,”Computing, vol. 107, no. 3, p. 82, 2025, https://doi.org/10.1007/s00607-025-01444-9

    work page doi:10.1007/s00607-025-01444-9 2025

  26. [26]

    Robust evaluation of gpu compute instances for hpc and ai in the cloud: a topsis approach with sensitivity, bootstrapping, and non-parametric analysis,

    M. Kumar, G. Kaur, and P. S. Rana, “Robust evaluation of gpu compute instances for hpc and ai in the cloud: a topsis approach with sensitivity, bootstrapping, and non-parametric analysis,”Computing, vol. 106, no. 12, pp. 3987–4014, 2024, https://doi.org/10.1007/s00607-024- 01342-6

    work page doi:10.1007/s00607-024- 2024

  27. [27]

    Evaluating hpc-style cpu performance and cost in virtualized cloud infras- tructures,

    J. Tharwani, S. Aggarwal, and A. A. Purkayastha, “Evaluating hpc-style cpu performance and cost in virtualized cloud infras- tructures,” inSoutheastCon 2025. IEEE, 2025, pp. 1239–1244, https://doi.org/10.1109/southeastcon56624.2025.10971705

    work page doi:10.1109/southeastcon56624.2025.10971705 2025

  28. [28]

    Adap- tive computing on the grid using apples,

    F. Berman, R. Wolski, H. Casanova, W. Cirne, H. Dail, M. Faer- man, S. Figueira, J. Hayes, G. Obertelli, J. Schopfet al., “Adap- tive computing on the grid using apples,”IEEE transactions on Par- allel and Distributed Systems, vol. 14, no. 4, pp. 369–382, 2003, https://doi.org/10.1109/tpds.2003.1195409

    work page doi:10.1109/tpds.2003.1195409 2003

  29. [29]

    In: Camara, O., et al

    B. Bethwaite, D. Abramson, F. Bohnert, S. Garic, C. Enticott, and T. Peachey, “Mixing grids and clouds: high-throughput science using the nimrod tool family,” inCloud computing: principles, systems and applications. Springer, 2010, pp. 219–237, https://doi.org/10.1007/978- 1-84996-241-4 13

    work page doi:10.1007/978- 2010

  30. [30]

    A taxonomy and sur- vey of grid resource management systems for distributed computing,

    K. Krauter, R. Buyya, and M. Maheswaran, “A taxonomy and sur- vey of grid resource management systems for distributed computing,” Software: Practice and Experience, vol. 32, no. 2, pp. 135–164, 2002, https://doi.org/10.1002/spe.432

    work page doi:10.1002/spe.432 2002

  31. [31]

    The evolution of the pegasus workflow management software,

    E. Deelman, K. Vahi, M. Rynge, R. Mayani, R. Ferreira da Silva, and G. Papadimitriou, “The evolution of the pegasus workflow management software,”Computing in Science & Engineering, vol. 21, no. 4, pp. 22– 36, 2019, https://doi.org/10.1109/MCSE.2019.2919690

    work page doi:10.1109/mcse.2019.2919690 2019

  32. [32]

    Scientific workflows and clouds,

    G. Juve and E. Deelman, “Scientific workflows and clouds,”XRDS: Crossroads, The ACM Magazine for Students, vol. 16, no. 3, pp. 14–18, 2010, https://doi.org/10.1145/1734160.1734166

    work page doi:10.1145/1734160.1734166 2010

  33. [33]

    Deadline based resource provisioning and scheduling algorithm for scientific workflows on clouds,

    M. A. Rodriguez and R. Buyya, “Deadline based resource provisioning and scheduling algorithm for scientific workflows on clouds,”IEEE Transactions on Cloud Computing, vol. 2, no. 2, pp. 222–235, 2014, https://doi.org/10.1109/tcc.2014.2314655

    work page doi:10.1109/tcc.2014.2314655 2014

  34. [34]

    arXiv preprint arXiv:2310.10634 , year=

    T. Xie, F. Zhou, Z. Cheng, P. Shi, L. Weng, Y . Liu, T. J. Hua, J. Zhao, Q. Liu, C. Liuet al., “Openagents: An open platform for language agents in the wild,”arXiv preprint arXiv:2310.10634, 2023, https://doi.org/10.48550/arXiv.2310.10634

    work page doi:10.48550/arxiv.2310.10634 2023

  35. [35]

    Auto-gpt for online decision making: Benchmarks and additional opinions.arXiv preprint arXiv:2306.02224,

    H. Yang, S. Yue, and Y . He, “Auto-gpt for online decision making: Benchmarks and additional opinions,”arXiv preprint arXiv:2306.02224, 2023, https://doi.org/10.48550/arXiv.2306.02224

    work page doi:10.48550/arxiv.2306.02224 2023

  36. [36]

    Autoagents: A framework for automatic agent generation

    G. Chen, S. Dong, Y . Shu, G. Zhang, J. Sesay, B. F. Karls- son, J. Fu, and Y . Shi, “Autoagents: A framework for auto- matic agent generation,”arXiv preprint arXiv:2309.17288, 2023, https://doi.org/10.48550/arXiv.2309.17288

    work page doi:10.48550/arxiv.2309.17288 2023

  37. [37]

    A Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications

    P. Sahoo, A. K. Singh, S. Saha, V . Jain, S. Mondal, and A. Chadha, “A systematic survey of prompt engineering in large language models: Techniques and applications,”arXiv preprint arXiv:2402.07927, vol. 1, 2024, https://doi.org/10.48550/arXiv.2402.07927

    work page internal anchor Pith review doi:10.48550/arxiv.2402.07927 2024

  38. [38]

    AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

    Q. Wu, G. Bansal, J. Zhang, Y . Wu, B. Li, E. Zhu, L. Jiang, X. Zhang, S. Zhang, J. Liuet al., “Autogen: Enabling next-gen llm applications via multi-agent conversations,” inFirst conference on language modeling, 2024, https://doi.org/10.48550/arXiv.2308.08155

    work page Pith review doi:10.48550/arxiv.2308.08155 2024

  39. [39]

    LLM Agents Making Agent Tools

    G. W ¨olflein, D. Ferber, D. Truhn, O. Arandjelovic, and J. N. Kather, “Llm agents making agent tools,” inProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2025, pp. 26 092–26 130, https://doi.org/10.18653/v1/2025.acl-long.1266

    work page doi:10.18653/v1/2025.acl-long.1266 2025

  40. [40]

    Evaluation and benchmark- ing of llm agents: A survey

    M. Mohammadi, Y . Li, J. Lo, and W. Yip, “Evaluation and benchmarking of llm agents: A survey,” inProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 2, 2025, pp. 6129–6139, https://doi.org/10.1145/3711896.3736570

    work page doi:10.1145/3711896.3736570 2025

  41. [41]

    Agent development kit (ADK),

    Google, “Agent development kit (ADK),” 2025, framework for building, evaluating, and deploying AI agents. [Online]. Available: https://github.com/google/adk-python

    2025

  42. [42]

    Least-to-Most Prompting Enables Complex Reasoning in Large Language Models

    D. Zhou, N. Sch ¨arli, L. Hou, J. Wei, N. Scales, X. Wang, D. Schuurmans, C. Cui, O. Bousquet, Q. Le, and E. Chi, “Least-to-most prompting enables complex reasoning in large language models,”arXiv preprint arXiv:2205.10625, 2022, https://doi.org/10.48550/arXiv.2205.10625

    work page internal anchor Pith review doi:10.48550/arxiv.2205.10625 2022

  43. [43]

    ReAct: Synergizing Reasoning and Acting in Language Models

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. R. Narasimhan, and Y . Cao, “React: Synergizing reasoning and acting in language models,” inThe eleventh international conference on learning representations, 2022, https://doi.org/10.48550/arXiv.2210.03629

    work page internal anchor Pith review doi:10.48550/arxiv.2210.03629 2022

  44. [44]

    Angr-the next generation of binary analysis,

    F. Wang and Y . Shoshitaishvili, “Angr-the next generation of binary analysis,” in2017 IEEE Cybersecurity Development (SecDev). IEEE, 2017, pp. 8–9, https://doi.org/10.1109/secdev.2017.14

    work page doi:10.1109/secdev.2017.14 2017

  45. [45]

    Eagle and K

    C. Eagle and K. Nance,The Ghidra Book: The Definitive Guide. no starch press, 2020, ISBN-13 978-1718501027

    2020

  46. [46]

    Pycg: Practical call graph generation in python,

    V . Salis, T. Sotiropoulos, P. Louridas, D. Spinellis, and D. Mitropoulos, “Pycg: Practical call graph generation in python,” in2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE). IEEE, 2021, pp. 1646–1657, https://doi.org/10.1109/icse43902.2021.00146

    work page doi:10.1109/icse43902.2021.00146 2021

  47. [47]

    Weisfeiler-lehman graph kernels,

    N. Shervashidze, P. Schweitzer, E. J. Van Leeuwen, K. Mehlhorn, and K. M. Borgwardt, “Weisfeiler-lehman graph kernels,”Journal of Machine Learning Research, vol. 12, no. 77, pp. 2539–2561, 2011, https://dl.acm.org/doi/10.5555/1953048.2078187

    work page doi:10.5555/1953048.2078187 2011

  48. [48]

    Turnbull,Monitoring with Prometheus

    J. Turnbull,Monitoring with Prometheus. Turnbull Press, 2018, ISBN- 13 978-0988820289

    2018

  49. [49]

    Node exporter,

    “Node exporter,” 2017, prometheus exporter for hardware and OS metrics exposed by *NIX kernels. [Online]. Available: https://github.com/prometheus/node exporter

    2017

  50. [50]

    Nvidia gpu exporter,

    “Nvidia gpu exporter,” 2021, nvidia GPU exporter for Prometheus. [On- line]. Available: https://github.com/utkuozdemir/nvidia gpu exporter

    2021

  51. [51]

    InProceedings of the 29th Symposium on Operating Systems Principles (SOSP ’23)

    W. Kwon, Z. Li, S. Zhuang, Y . Sheng, L. Zheng, C. H. Yu, J. E. Gonzalez, H. Zhang, and I. Stoica, “Efficient memory management for large language model serving with pagedattention,” inProceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles, 2023, pp. 611–626, https://doi.org/10.1145/3600006.3613165

    work page doi:10.1145/3600006.3613165 2023

  52. [52]

    Hpcg benchmark technical specification,

    M. A. Heroux, J. Dongarra, and P. Luszczek, “Hpcg benchmark technical specification,” Sandia National Laboratories (SNL-NM), Albuquerque, NM (United States), Tech. Rep., 2013, https://doi.org/10.2172/1113870

    work page doi:10.2172/1113870 2013

  53. [53]

    IOzone filesystem benchmark,

    W. D. Norcott, “IOzone filesystem benchmark,” 2003, https://www.iozone.org/

    2003

  54. [54]

    A benchmark suite for improving per- formance portability of the sycl programming model,

    Z. Jin and J. S. Vetter, “A benchmark suite for improving per- formance portability of the sycl programming model,” in2023 IEEE International Symposium on Performance Analysis of Sys- tems and Software (ISPASS). IEEE, 2023, pp. 325–327, https://doi.org/10.1109/ISPASS57527.2023.00041

    work page doi:10.1109/ispass57527.2023.00041 2023

  55. [55]

    LAMMPS-a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales,

    A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. In’t Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyenet al., “Lammps-a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales,”Computer physics communications, vol. 271, p. 108171, 2022, https://doi.org/10.1016/j.c...

    work page doi:10.1016/j.cpc.2021.108171 2022

  56. [56]

    Available: https://doi.org/10.2172/1090032

    I. Karlin, J. Keasler, and J. R. Neely, “Lulesh 2.0 updates and changes,” Lawrence Livermore National Laboratory (LLNL), Livermore, CA (United States), Tech. Rep., 2013, https://doi.org/10.2172/1090032

    work page doi:10.2172/1090032 2013

  57. [57]

    Palabos: parallel lattice boltzmann solver,

    J. Latt, O. Malaspinas, D. Kontaxakis, A. Parmigiani, D. La- grava, F. Brogi, M. B. Belgacem, Y . Thorimbert, S. Leclaire, S. Liet al., “Palabos: parallel lattice boltzmann solver,”Comput- ers & Mathematics with Applications, vol. 81, pp. 334–350, 2021, https://doi.org/10.1016/j.camwa.2020.03.022

    work page doi:10.1016/j.camwa.2020.03.022 2021

  58. [58]

    Accelerating financial applications on the gpu,

    S. Grauer-Gray, W. Killian, R. Searles, and J. Cavazos, “Accelerating financial applications on the gpu,” inProceedings of the 6th Workshop on General Purpose Processor Using Graphics Processing Units, 2013, pp. 127–136, https://dx.doi.org/10.1145/2458523.2458536

    work page doi:10.1145/2458523.2458536 2013

  59. [59]

    LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code

    N. Jain, K. Han, A. Gu, W.-D. Li, F. Yan, T. Zhang, S. Wang, A. Solar-Lezama, K. Sen, and I. Stoica, “Livecodebench: Holistic and contamination free evaluation of large language models for code,” in The Thirteenth International Conference on Learning Representations, 2025, https://doi.org/10.48550/arXiv.2403.07974

    work page internal anchor Pith review doi:10.48550/arxiv.2403.07974 2025

  60. [60]

    BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions

    T. Y . Zhuo, M. C. Vu, J. Chim, H. Hu, W. Yu, R. Widyasari, I. N. B. Yusuf, H. Zhan, J. He, I. Paul, S. Brunner, C. Gong, J. Hoang, A. R. Zebaze, X. Hong, W.-D. Li, J. Kaddour, M. Xu, Z. Zhang, P. Yadav, N. Jain, A. Guet al., “Bigcodebench: Benchmarking code generation with diverse function calls and complex instructions,” inThe Thirteenth International C...

    work page internal anchor Pith review doi:10.48550/arxiv.2406.15877 2025

  61. [61]

    How efficient is llm-generated code? a rigorous & high-standard benchmark,

    R. Qiu, W. Zeng, J. Ezick, C. Lott, and H. Tong, “How efficient is llm-generated code? a rigorous & high-standard benchmark,” inThe Thirteenth International Conference on Learning Representations, 2025, https://doi.org/10.48550/arXiv.2406.06647

    work page doi:10.48550/arxiv.2406.06647 2025