Pith. sign in

REVIEW 3 major objections 4 minor 42 references

A 31-billion-parameter open multimodal model, fine-tuned on distilled chain-of-thought rationales and paired with BM25 retrieval over nuclear handbooks, meets the regulatory 80% passing criterion on 8 of 14 U.S. Nuclear Regulatory Commissio

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

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

T0 review · deepseek-v4-flash

2026-08-01 05:52 UTC pith:NAQNS6QX

load-bearing objection A mostly solid new benchmark study that is honest about its own limits, with one unresolved data-leakage risk that could shift the headline 8/14 pass count. the 3 major comments →

arxiv 2607.22067 v1 pith:NAQNS6QX submitted 2026-07-24 cs.CL cs.AI

Benchmarking Fine-tuning and Retrieval Strategies for a Multimodal Language Model on the NRC Reactor Operator Licensing Examination

Isak Hwang , Yoon Pyo Lee This is my paper
classification cs.CL cs.AI
keywords Nuclear EngineeringReactor Operator LicensingGeneric Fundamentals ExaminationLarge Language ModelsMultimodal ModelsRetrieval-Augmented GenerationSupervised Fine-TuningChain-of-Thought Distillation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper asks whether an open, 31-billion-parameter multimodal model can reach operator-level command of nuclear engineering fundamentals. It answers that supervised fine-tuning on chain-of-thought rationales distilled from a stronger teacher, combined with BM25 retrieval over Department of Energy Fundamentals Handbook chunks, lets the model meet the NRC's strict 80% passing criterion on 8 of 14 administered Generic Fundamentals Examinations, while no configuration without fine-tuning passes any. This matters because the evaluation uses the exact instrument and threshold applied to human operator candidates, and the entire pipeline runs on a single workstation with open components. The paper also reports that the best chunking strategy reverses after fine-tuning and that retrieval-augmented fine-tuning underperforms plain supervised fine-tuning under matched conditions.

Core claim

On the paper's own terms, the central discovery is that distilled chain-of-thought fine-tuning plus BM25 retrieval over fixed-size sliding-window chunks of the DOE Fundamentals Handbook lifts an open 31B multimodal model from 51.86% pooled accuracy (0/14 papers passed) to 79.66% pooled accuracy with 8 of 14 papers meeting the 80% criterion; the PWR item set alone reaches 80.23%. The authors stress that the per-examination pass count, not pooled accuracy, is the primary result because the regulatory threshold applies to single administrations, and they note the confidence interval spans 80%, so the result is properly read as approaching operator-level command rather than reliably achieving it

What carries the argument

The load-bearing machinery is the pipeline: supervised fine-tuning with LoRA on 3,577 question-rationale pairs, where the rationales are generated by a proprietary teacher instructed to reason step-by-step and end with 'Therefore, the correct answer is (X)'; BM25 sparse lexical retrieval over the seven-volume DOE Fundamentals Handbook, segmented either by fixed-size sliding windows (1,000 characters, 200 overlap) or by structure-aware typographic section chunks; and the evaluation unit itself, the per-examination pass indicator Pass(e)=1 when correct/total ≥ 0.80, mirroring the NRC's no-rounding rule. The interplay of chunking strategy and model training state is what the paper claims determ

Load-bearing premise

The load-bearing assumption is that text-based deduplication removed essentially all overlap between the training question banks and the 700 evaluation items; the paper itself estimates up to 88 evaluation questions may still have identical bank counterparts in training, and if those leaked items cluster in the exams that reached the 80% threshold, the 8/14 pass count would be inflated by memorization rather than generalization.

What would settle it

Take the 228 evaluation questions whose normalized text had no match anywhere in the question banks (the items the NRC says are newly authored or heavily derived) and re-score the fourteen exams using only those items. If the pass count drops sharply or vanishes, the reported 8/14 result is largely memorization of the bank; if it holds, the model genuinely generalizes. A cheaper version is to identify the 88 potentially leaked items and check whether they are over-represented in the exams that met the 80% mark.

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

If this is right

  • If correct, a 31B open-weight model can be brought from failing the NRC GFE outright (51.9% pooled, 0/14 passes) to meeting the 80% criterion on 8 of 14 administrations using only fine-tuning and sparse retrieval, with no larger model or architecture change.
  • No configuration without fine-tuning passed any exam, so retrieval alone is insufficient; domain-adaptation of the model is a necessary ingredient for this task.
  • Chunking strategy must be re-tuned after fine-tuning: the preferred strategy flips from structure-aware (base model) to fixed-size (fine-tuned), so retrieval design and training cannot be decoupled.
  • RAFT underperforms plain SFT by about 2.2–2.3 points under matched chunking and retrieval; the paper attributes this to the teacher's dense-retriever context differing from the student's sparse-retriever context and proposes a passage-overlap measurement as the decisive follow-up.
  • The entire fine-tuning and inference pipeline runs on one commodity workstation with publicly available data, meaning the deployed artifact needs no proprietary model or network at runtime; the only external dependency is the single offline distillation step.

Where Pith is reading between the lines

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

  • The paper's own estimate that up to 88 of 700 evaluation items may still have bank counterparts in the training corpus means the 8/14 pass count could be partially memorization-driven; re-scoring only the 228 items with no textual bank match would test whether the result survives strict generalization.
  • The per-examination versus pooled-accuracy reversal (PWR passes 3/7 with 80.23% pooled; BWR passes 5/7 with 79.08% pooled) suggests that any LLM benchmark inheriting an external passing criterion should also inherit the criterion's unit of assessment; average accuracy can invert qualitative conclusions, a lesson likely transferable to medical and legal licensing benchmarks.
  • The chunking-reversal result implies that RAG hyperparameters tuned on an off-the-shelf model may be actively suboptimal after fine-tuning; teams building domain-adapted RAG systems should co-tune chunking with the fine-tuning stage rather than inheriting base-model settings.
  • If the RAFT deficit is indeed caused by teacher/student retriever mismatch, then regenerating the distilled rationales with the same sparse retriever used at inference—instead of the dense retriever used by the teacher—could flip the ordering; this is a cheap, testable fix that the paper does not run.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper benchmarks a 31B open-weight multimodal model (Gemma 4 31B-IT) on the NRC Reactor Operator Generic Fundamentals Examination, using the regulatory 80% passing criterion. It compares eight configurations formed by crossing model state (base, SFT, RAFT) with retrieval condition (none, BM25 RAG over DOE handbooks with fixed-size or structure-aware chunking). The headline result is that SFT with fixed-size chunking RAG passes 8 of the 14 March 2015–2021 examinations, while no configuration without fine-tuning passes any; pooled accuracy is 79.66% (95% Wilson CI spanning the threshold) and PWR-only accuracy is 80.23%. The paper also reports a chunking-strategy reversal with training state and a small but consistent RAFT deficit relative to SFT.

Significance. If the pass count is taken at face value, the paper provides an existence proof that a moderately large open-weight multimodal model, adapted on a single workstation with distilled chain-of-thought supervision and BM25 retrieval, can approach a regulatory knowledge bar on a majority of administered exams. The evaluation is unusually transparent: the paper states the exact per-examination margins, presents both scoring conventions, and explicitly acknowledges the residual overlap between training banks and evaluation items. The public data sources make the core pipeline independently replicable in principle, although code and artifacts are not released. The secondary claims—chunking reversal and RAFT underperformance—are less secure because they rest on small pooled differences and a known asymmetry in the RAFT training data, respectively. The paper is honest about these weaknesses, but the main 8/14 claim remains vulnerable to the residual data-leakage risk it identifies.

major comments (3)
  1. [§2.2.2, §4.7, Table 7] The residual training/evaluation overlap is load-bearing for the central claim. Deduplication by normalized text removes only verbatim matches, yet NRC construction implies only 140 of 700 evaluation items should lack a bank counterpart; the paper reports 228 unmatched, leaving up to 88 items whose bank counterparts may remain in the 3,577-item training corpus (§4.7). Several passing cells in Table 7 sit exactly at the 80.0% boundary (2019 PWR, 2020 BWR, 2021 BWR are 40/50; 2018 BWR is 41/50), and the PWR pooled figure clears the criterion by one item (280/349). A single memorized leaked item on any of these boundary examinations flips the per-exam pass indicator, changing the headline from 8/14 to as low as 5/14. The paper calls this a 'bounded but non-zero source of optimism' but does not quantify its effect on Table 4. The authors should provide a sensitivity analysis, e.g., conservat
  2. [§2.2.3, §3.5, §4.4] The claim that RAFT underperforms SFT 'in matching search environments' is not supported by the experimental design. As the paper states in §2.2.3, the RAFT training examples pair a sparse-retrieved context block with a rationale authored against dense-retrieved passages; the supervision signal may cite evidence absent from the accompanying context. Section 3.5 nevertheless concludes that the deficit 'is attributable to retrieval conditioning during fine-tuning,' and the abstract and conclusion repeat the RAFT-underperformance finding. This attribution is confounded by the teacher-student retriever mismatch. The paper itself identifies the passage-overlap measurement as the primary follow-up (§4.4). Until that measurement is reported, or the claim is explicitly downgraded to a hypothesis, the RAFT conclusion is overstated.
  3. [§3.1, §3.7, Table 4] The primary metric is the number of examinations passed, and the paper calls the top configuration 'the strongest' and 'outperforming all alternatives.' With 14 binary outcomes, an 8-versus-5 difference among fine-tuned configurations is within sampling variability; the paper provides no interval or paired test on the pass count. Section 3.7 explicitly disclaims statistically significant separation among fine-tuned configurations from the pooled figures, but the pass-count claim is presented without a corresponding uncertainty quantification. The authors should either provide a paired test (e.g., McNemar on per-exam pass/fail status between configurations) or qualify the 'outperforming' language as descriptive rather than inferential.
minor comments (4)
  1. [§2.2.2] The text says '3,578 items were retained' after deduplication, while Table 1 and §2.2.3 report 3,577 after one item was discarded during rationale generation. Consider making the two-step arithmetic explicit in one place.
  2. [§4.7] The paper reports that 121 training items have fewer than four parsed answer options and 108 have none, and that these were used for fine-tuning. This is a notable data-quality issue; please state how many such items are in the final training corpus after any filtering, and discuss any impact on the SFT/RAFT results.
  3. [§4.3] The chunking-strategy reversal is presented as a robust regularity, but the differences (62.18 vs 64.47 for base; 79.66 vs 77.51 for SFT) are within the overlapping Wilson intervals, and the PWR RAFT pair is exactly tied. Consider labeling this a descriptive pattern rather than a mechanistic finding, or provide a significance test.
  4. [Data availability] The paper states that prompts, chunking parameters, and evaluation scripts are 'available from the corresponding author on reasonable request.' Given the paper's emphasis on reproducibility, a permanent public repository would strengthen the contribution; the current arrangement makes independent verification contingent on author response.

Circularity Check

1 steps flagged

Benchmark is partially self-referential: the training corpus is the same NRC question-bank pool from which the evaluation exams are constructed, and exact-text deduplication still leaves up to 88 evaluation items with probable bank counterparts in training.

specific steps
  1. fitted input called prediction [§2.2.1–2.2.2, §3 (Results opening), §4.7 (Limitations)]
    "The first is the pair of NRC GFE question banks ... the authoritative pools from which examination questions are drawn. Both are used to construct the training corpus. ... Forty items are drawn directly from the bank. ... Reported scores therefore measure generalization to unseen questions rather than recall of the question bank. ... The excess of 88 bounds the number of evaluation questions whose bank counterparts may remain in the training corpus, or roughly 2.5% of it."

    The model is fine-tuned on the same NRC question banks from which, by the paper's own description of NRC construction, 40 of every 50 evaluation items are drawn and 5 more are derived. Deduplication removes only exact normalized-text matches, leaving up to 88 evaluation items that may still have bank counterparts in the 3,577-item training set. The phrase 'generalization to unseen questions' is therefore true only under the paper's exact-match definition of 'unseen,' not under the NRC's bank-sampling construction. Since several passes sit exactly at 80% (2019 PWR, 2020 BWR, 2021 BWR at 40/50), the residual bank overlap could materially inflate the 8/14 pass count. The paper calls this a 'bounded but non-zero source of optimism' but does not re-analyze Table 4 under that bound.

full rationale

There is no equation-level circularity: the SFT objective (Eq. 3), RAFT objective (Eq. 4), BM25 scoring (Eqs. 5–6), and the regulatory pass criterion (Eq. 1) are all defined independently of the reported outcomes, and no uniqueness theorem or load-bearing self-citation is used. The only author-overlapping citation ([12], Y.P. Lee et al.) is contextual background, not a load-bearing premise. The central circularity-style concern is data-level: the evaluation set is a set of exams drawn from the same public NRC question banks that formed the fine-tuning corpus, and exact-text deduplication is the only separator. The paper itself quantifies a residual overlap of up to 88 evaluation items with plausible bank counterparts, which directly threatens the interpretation that the 8/14 result measures generalization rather than partial memorization. This is a genuine but partial circularity/contamination issue, not a derivation that is equivalent to its inputs by construction; the model still must answer newly authored and modified items, and the comparison among eight configurations has independent empirical content. A moderate score of 4 reflects that partial data-level circularity rather than a full equational or self-citation collapse.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

The paper is an empirical benchmark rather than a derivation, so the ledger lists the hand-chosen hyperparameters and methodological assumptions that the central claim depends on. No new physical or mathematical entities are introduced. The most load-bearing entries are the deduplication assumption and the teacher-rationale correctness assumption, both acknowledged by the authors as imperfect.

free parameters (6)
  • Fixed-size chunk size and overlap = N_size=1000 chars, N_overlap=200 chars
    Hand-chosen with no ablation; central to the RAG configurations and the chunking-reversal claim.
  • Retrieval top-k and context budget = top 4 chunks, 7000 characters
    Hand-chosen; no sensitivity analysis reported.
  • BM25 constants = k1=1.5, b=0.75
    Standard defaults, not tuned to the nuclear domain.
  • LoRA hyperparameters = r=16, alpha=32, dropout=0.05, epochs=3, lr=2e-4
    Hand-chosen without reported grid search; affects the SFT/RAFT results.
  • Structure-aware heading detection threshold = font size >= f_base + 0.8pt plus TOC/numbered-prefix conditions
    Heuristic thresholds for heading detection; could change chunk boundaries and retrieval behavior.
  • Minimum indexed chunk length = 80 characters
    Arbitrary cutoff to drop headers/page artefacts; affects which chunks can be retrieved.
axioms (5)
  • domain assumption The NRC GFE items and DOE handbook texts were correctly downloaded, parsed, and rendered, including figures.
    If PDF parsing or image extraction is wrong, scores are not valid; no independent verification is provided.
  • domain assumption Gemini 3 Flash distilled rationales are sufficiently correct to serve as training targets.
    The student inherits teacher errors; no faithfulness or accuracy audit of the rationales is reported.
  • domain assumption Deduplication by normalized question text removes essentially all training–evaluation overlap.
    The paper itself qualifies this as incomplete: 'bounded but non-zero source of optimism' (§4.7).
  • domain assumption The cascaded answer extractor maps free-form generations to the intended option without systematic bias.
    Extraction accuracy is not manually audited beyond the two scoring conventions.
  • domain assumption The eight configurations differ only in the stated factors.
    RAFT is confounded by the teacher/student retriever mismatch (dense vs sparse), as acknowledged in §2.2.3 and §4.4.

pith-pipeline@v1.3.0-alltime-deepseek · 19811 in / 12130 out tokens · 121122 ms · 2026-08-01T05:52:06.425677+00:00 · methodology

0 comments
read the original abstract

The integration of large language models (LLMs) into the nuclear power industry requires outputs grounded in domain-specific knowledge. This study evaluates a 31-billion-parameter open-weight multimodal model (Gemma 4 31B-IT) on its capacity to apply nuclear knowledge by benchmarking eight model-retrieval configurations against the U.S. Nuclear Regulatory Commission (NRC) Reactor Operator licensing examination. We evaluate 14 Generic Fundamentals Examinations (GFE) from the 2015-2021 March sittings (seven pressurized and seven boiling water reactor exams) using the standard 80% human passing criterion. The base model is compared against configurations utilizing supervised fine-tuning (SFT) on Gemini-distilled chain-of-thought (CoT) rationales, retrieval-augmented generation (RAG) with BM25 sparse retrieval over the U.S. Department of Energy Fundamentals Handbook, and retrieval-augmented fine-tuning (RAFT). Within the retrieval pipeline, we compare fixed-size sliding-window chunking against structure-aware chunking. The SFT configuration with fixed-size chunking RAG met the criterion on 8 of the 14 examinations, outperforming all alternatives, whereas no configuration without fine-tuning passed any. Aggregate accuracy reached 79.7%, with a confidence interval spanning the threshold, and 80.2% on PWR items specifically. Furthermore, two regularities emerged: the preferred chunking strategy reverses depending on the model's training state, and RAFT underperforms compared to standard SFT in matching search environments. These results demonstrate which combination of fine-tuning and search approaches achieves operator-level capabilities.

Figures

Figures reproduced from arXiv: 2607.22067 by Isak Hwang, Yoon Pyo Lee.

Figure 1
Figure 1. Figure 1: Pipeline of the proposed framework. The question banks and the administered papers are obtained independently from the NRC and the papers overlap the banks. Deduplication removes from the pooled banks every item whose normalized question text matches an evaluation item, leaving 3,577 training questions that then receive distilled CoT rationales and are used for low-rank fine-tuning. The DOE handbook corpus… view at source ↗

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

42 extracted references · 2 linked inside Pith

  1. [1]

    Pakarinen, M

    S. Pakarinen, M. Sallinen, Workload management measures for sup- portingnuclearindustrymaincontrolroomoperatorsandemergency response organization personnel during crises—a scoping review, Industrial health 63 (2025) 214–241

    2025

  2. [2]

    Nuclear Regulatory Commission, Washington, DC, 2025

    U.S.NuclearRegulatoryCommission,AHumanErrorHandbookand Verification of IDHEAS-DATA, Research Information Letter 2025- 02, U.S. Nuclear Regulatory Commission, Washington, DC, 2025

    2025

  3. [3]

    U.S. Nuclear Regulatory Commission, Title 10, code of federal regulations, part 55—Operators’ Licenses, Electronic Code of Fed- eralRegulations(eCFR),2024.URL:https://www.ecfr.gov/current/ title-10/chapter-I/part-55, sections 55.40 and 55.41

    2024

  4. [4]

    Nuclear Regulatory Commission, Operator Licensing Exami- nation Standards for Power Reactors, Final Report NUREG-1021, Rev

    U.S. Nuclear Regulatory Commission, Operator Licensing Exami- nation Standards for Power Reactors, Final Report NUREG-1021, Rev. 12, U.S. Nuclear Regulatory Commission, Office of Nuclear Reactor Regulation, Washington, DC, USA, 2021

    2021

  5. [5]

    Singhal, S

    K. Singhal, S. Azizi, T. Tu, S. S. Mahdavi, J. Wei, H. W. Chung, N.Scales,A.Tanwani,H.Cole-Lewis,S.Pfohl,etal., Largelanguage models encode clinical knowledge, Nature 620 (2023) 172–180

    2023

  6. [6]

    H. Nori, N. King, S. M. McKinney, D. Carignan, E. Horvitz, Ca- pabilities of GPT-4 on medical challenge problems, arXiv preprint arXiv:2303.13375 (2023)

    Pith/arXiv arXiv 2023

  7. [7]

    D. M. Katz, M. J. Bommarito, S. Gao, P. Arredondo, GPT-4 passes thebarexam, PhilosophicalTransactionsoftheRoyalSocietyA382 (2024) 20230254

    2024

  8. [8]

    C. Wang, D. Mandelli, J. Cogliati, Technical language processing of nuclear power plants equipment reliability data, Energies 17 (2024) 1785

    2024

  9. [9]

    L. Wang, J. Hwang, K. Han, A. Gupta, Large language model- driven workflow for regulatory reporting in nuclear construction, in: TransactionsoftheInternationalConferenceonStructuralMechanics in Reactor Technology (SMiRT), IASMiRT, 2025

    2025

  10. [10]

    B. Qi, X. Xiao, J. Liang, H. Zhou, W. Wu, X. Zhang, Synergistic ai fornuclearsafety:Pinns,xai,dts,andllmsinfaultdiagnosis, Nuclear Technology (2026) 1–18

    2026

  11. [11]

    Fayyaz, W

    Y. Fayyaz, W. Elouataoui, Y. Gahi, K. El-Khatib, G. Harvel, K. Sankaranarayanan, Natural language processing in the nuclear industry: Opportunities and challenges, Nuclear Technology (2025) 1–21

    2025

  12. [12]

    Y. P. Lee, J. Cha, Y. Yu, S. G. Kim, Large language model agent for nuclear reactor operation assistance, Nuclear Engineering and Technology 57 (2025) 103842

    2025

  13. [13]

    L.Lin,T.Reeves,V.Agarwal,UsingaLargeLanguageModelforAc- curateTechnicalLanguageGenerationinthePredictiveMaintenance of Circulating Water Systems in Nuclear Power Plants, Technical Report, Idaho National Laboratory (INL), Idaho Falls, ID (United States), 2025

    2025

  14. [14]

    Hsieh, C.-L

    C.-Y. Hsieh, C.-L. Li, C.-K. Yeh, H. Nakhost, Y. Fujii, A. Ratner, R.Krishna,C.-Y.Lee,T.Pfister, Distillingstep-by-step!outperform- ing larger language models with less training data and smaller model sizes, in: Findings of the Association for Computational Linguistics: ACL 2023, Association for Computational Linguistics, 2023, pp. 8003–8017. doi:10.18653...

    doi:10.18653/v1/2023.findings-acl.507 2023

  15. [15]

    L. C. Magister, J. Mallinson, J. Adamek, E. Malmi, A. Severyn, Teachingsmalllanguagemodelstoreason,in:Proceedingsofthe61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), 2023, pp. 1773–1781

    2023

  16. [16]

    Lewis, E

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V. Karpukhin, N. Goyal, H. Küttler, M. Lewis, W.-t. Yih, T. Rocktäschel, S. Riedel, D. Kiela, Retrieval-augmented generation for knowledge-intensive NLP tasks, I. Hwang and Y.P. Lee:Preprint submitted to ElsevierPage 13 of 14 Benchmarking LLMs on NRC Reactor Operator Licensing Exams in: Advances in Neural Informa...

    2020

  17. [17]

    Zhang, S

    T. Zhang, S. G. Patil, N. Jain, S. Shen, M. Zaharia, I. Stoica, J. E. Gonzalez, RAFT:AdaptinglanguagemodeltodomainspecificRAG, arXiv preprint arXiv:2403.10131 (2024)

    Pith/arXiv arXiv 2024

  18. [18]

    LlamaIndex Team, Evaluating the ideal chunk size for a RAG system using LlamaIndex,https://www.llamaindex.ai/blog/ evaluating-the-ideal-chunk-size-for-a-rag-system-using-llamaindex-6207e5d3fec5,

  19. [19]

    U.S. Department of Energy, DOE Fundamentals Handbook: Thermodynamics, Heat Transfer, and Fluid Flow, Volumes 1–3, Handbook DOE-HDBK-1012/1-92, DOE-HDBK-1012/2-92, DOE-HDBK-1012/3-92, U.S. Department of Energy, Washington, DC, 1992

    1992

  20. [20]

    Department of Energy, DOE Fundamentals Handbook: In- strumentation and Control, Volumes 1–2, Handbook DOE-HDBK- 1013/1-92, DOE-HDBK-1013/2-92, U.S

    U.S. Department of Energy, DOE Fundamentals Handbook: In- strumentation and Control, Volumes 1–2, Handbook DOE-HDBK- 1013/1-92, DOE-HDBK-1013/2-92, U.S. Department of Energy, Washington, DC, 1992

    1992

  21. [21]

    Department of Energy, DOE Fundamentals Handbook: Nu- clear Physics and Reactor Theory, Volumes 1–2, Handbook DOE- HDBK-1019/1-93, DOE-HDBK-1019/2-93, U.S

    U.S. Department of Energy, DOE Fundamentals Handbook: Nu- clear Physics and Reactor Theory, Volumes 1–2, Handbook DOE- HDBK-1019/1-93, DOE-HDBK-1019/2-93, U.S. Department of En- ergy, Washington, DC, 1993

    1993

  22. [22]

    Accessed: June 2026

    Gemma Team, Google DeepMind, Gemma 4 model card,https:// ai.google.dev/gemma/docs/core/model_card_4, 2026. Accessed: June 2026

    2026

  23. [23]

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination question bank — pressurized water reactor,https://www.nrc.gov/reactors/operator-licensing/ history-rulemaking-activities/generic-fundamentals-examinations/ pwr, 2020. 2,140 questions organized by topic. Accessed: June 2026

    2020

  24. [24]

    Nuclear Regulatory Commission, NRC generic fundamentals examinationquestionbank—boilingwaterreactor,https://www.nrc

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examinationquestionbank—boilingwaterreactor,https://www.nrc. gov/reactors/operator-licensing/history-rulemaking-activities/ generic-fundamentals-examinations/bwr, 2020. 2,149 questions organized by topic. Accessed: June 2026

    2020

  25. [25]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2015,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2015,https://www. nrc.gov/docs/ML1617/ML16173A029.pdf, 2015. ADAMS Accession No. ML16173A029

    2015

  26. [26]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2016,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2016,https://www. nrc.gov/docs/ML1630/ML16300A005.pdf, 2016. ADAMS Accession No. ML16300A005

    2016

  27. [27]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2017,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2017,https://www. nrc.gov/docs/ML1807/ML18074A026.pdf, 2017. ADAMS Accession No. ML18074A026

    2017

  28. [28]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2018,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2018,https://www. nrc.gov/docs/ML1826/ML18269A212.pdf, 2018. ADAMS Accession No. ML18269A212

    2018

  29. [29]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2019,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2019,https://www. nrc.gov/docs/ML1929/ML19290H601.pdf, 2019. ADAMS Accession No. ML19290H601

    2019

  30. [30]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2020,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2020,https://www. nrc.gov/docs/ML2110/ML21103A276.pdf, 2020. ADAMS Accession No. ML21103A276

    2020

  31. [31]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2021,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — pressurized water reactor, march 2021,https://www. nrc.gov/docs/ML2200/ML22007A311.pdf, 2021. ADAMS Accession No. ML22007A311

    2021

  32. [32]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2015,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2015,https://www. nrc.gov/docs/ML1617/ML16173A025.pdf, 2015. ADAMS Accession No. ML16173A025

    2015

  33. [33]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2016,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2016,https://www. nrc.gov/docs/ML1630/ML16300A003.pdf, 2016. ADAMS Accession No. ML16300A003

    2016

  34. [34]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2017,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2017,https://www. nrc.gov/docs/ML1807/ML18074A024.pdf, 2017. ADAMS Accession No. ML18074A024

    2017

  35. [35]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2018,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2018,https://www. nrc.gov/docs/ML1826/ML18269A207.pdf, 2018. ADAMS Accession No. ML18269A207

    2018

  36. [36]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2019,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2019,https://www. nrc.gov/docs/ML1929/ML19290H604.pdf, 2019. ADAMS Accession No. ML19290H604

    2019

  37. [37]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2020,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2020,https://www. nrc.gov/docs/ML2110/ML21103A277.pdf, 2020. ADAMS Accession No. ML21103A277

    2020

  38. [38]

    Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2021,https://www

    U.S. Nuclear Regulatory Commission, NRC generic fundamentals examination — boiling water reactor, march 2021,https://www. nrc.gov/docs/ML2200/ML22007A312.pdf, 2021. ADAMS Accession No. ML22007A312

    2021

  39. [39]

    U.S. Nuclear Regulatory Commission, Generic fundamentals examinations for reactor operators,https://www.nrc.gov/ reactors/operator-licensing/history-rulemaking-activities/ generic-fundamentals-examinations, 2022. Page last reviewed 31 March 2022

    2022

  40. [40]

    E. J. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, W. Chen, LoRA: Low-rank adaptation of large language models, in: International Conference on Learning Representations, OpenRe- view.net, 2022. URL:https://openreview.net/forum?id=nZeVKeeFYf9

    2022

  41. [41]

    S.E.Robertson,H.Zaragoza, Theprobabilisticrelevanceframework: BM25 and beyond, Foundations and Trends in Information Retrieval 3 (2009) 333–389

    2009

  42. [42]

    Accessed: June 2026

    A.Hannun,J.Digani,A.Katharopoulos,R.Collobert,MLX:Efficient and flexible deep learning on Apple Silicon,https://github.com/ ml-explore/mlx, 2023. Accessed: June 2026. I. Hwang and Y.P. Lee:Preprint submitted to ElsevierPage 14 of 14

    2023