Pith. sign in

REVIEW 4 major objections 8 minor 26 references

A retrieval-augmented pipeline that feeds expert causal triples into a fine-tuned language encoder can classify hidden bridge damage causes at 87% accuracy, and a 4-bit quantized version does it with 72% less memory while generalizing bette

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 07:51 UTC pith:VN4DSYBP

load-bearing objection Honest engineering paper with a useful pipeline and a clean comparison setup, but the Golden Testset appears contaminated by retrieval leakage from the index, so the headline equal-accuracy and generalization claims need rework. the 4 major comments →

arxiv 2607.21680 v1 pith:VN4DSYBP submitted 2026-07-23 cs.LG

Encoding Invisible Causation for Bridge Diagnostic Agents: Triple-Guided Retrieval-Augmented Fine-Tuning with QLoRA

Takato Yasuno This is my paper
classification cs.LG
keywords bridge inspectiondamage cause classificationcausal knowledge triplesretrieval-augmented fine-tuningparameter-efficient fine-tuningquantizationGolden Testsetdiagnostic agents
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 claims that a bridge inspection agent can learn to infer invisible damage causes—salt intrusion, frost damage, fatigue cracking, and seven others—by retrieving cause-effect statements extracted from maintenance manuals and appending them to the visible damage description before classification. The core mechanism is a causal triple, a structured statement like 'rebar corrosion caused by chloride ion concentration exceeding 1.2 kg/m³', thousands of which are harvested from 15–35 PDF manuals and indexed for similarity search. At both training and inference, relevant triples are retrieved for each input, converting tacit expert knowledge into explicit text the encoder can use. The authors then compare three fine-tuning strategies and find that a 4-bit quantized version matches the full-precision model's 87% accuracy on a controlled 116-sample test set while using 72% less GPU memory, and it scores 13 points higher on 100 diverse unseen inputs. The significance is practical: memory-efficient diagnostic agents could run on consumer-grade hardware for edge deployment, and the quantized variant appears to generalize better, not just compress.

Core claim

On a fixed, stratified Golden Testset of 116 samples, the 4-bit quantized fine-tuning variant reaches exactly the same accuracy as full-precision low-rank fine-tuning—87.07%—with nearly identical weighted F1 (0.869 vs 0.870), while cutting inference time by 11% and GPU memory from 1.45 GB to 0.40 GB. More strikingly, on a separate 100-sample evaluation spanning all 10 damage-cause classes, the quantized model classifies 47.0% correctly versus 34.0% for the full-precision model. The paper interprets this 13-point gap as evidence that the discretization noise of 4-bit quantization acts as a regularizer, preventing the model from memorizing training idiosyncrasies and improving generalization u

What carries the argument

The causal knowledge triple—a structured (subject, relation, object) assertion such as '(rebar corrosion, caused by, chloride ion concentration exceeding 1.2 kg/m³)'—is the unit of expert knowledge. A large language model extracts 6,745 such triples from 35 diagnostic PDF manuals; each is embedded in a dense vector space and stored in a similarity-search index. For each damage description, the top-k most similar triples are retrieved, filtered by a LLM relevance judgment, and then concatenated onto the description to form the encoder's input. This retrieval-augmented context is the same at training and inference, which is what converts tacit, invisible causal knowledge into explicit signal t

Load-bearing premise

The entire generalization claim rests on the 100 diverse unseen inputs being a fair, representative sample of real deployment text; they were hand-picked by the authors, are not publicly released, and no independent labeling protocol is described, so a biased or non-representative sample could make the 13-point advantage an artifact.

What would settle it

Publish the Golden Testset and the 100 unseen inputs, have independent inspectors write a fresh set of damage descriptions sourced from manual blocks that were deliberately withheld from the triple store, and run both fine-tuned variants on them. If the 4-bit model no longer matches or beats the full-precision model on that fresh set, the paper's central recommendation—that the quantized variant is the optimal fine-tuning strategy—collapses.

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

If this is right

  • Bridge inspection agencies can run cause-diagnosis on local consumer GPUs (0.40 GB memory) instead of renting cloud accelerators, enabling on-premise edge deployment.
  • The stratified, deduplicated Golden Testset gives future research a fixed benchmark for comparing fine-tuning strategies on the same 116 samples, making results comparable across studies.
  • Quantization as implicit regularization suggests that for small, balanced datasets, 4-bit quantized fine-tuning may outperform full-precision fine-tuning—a principle that could transfer to other classification domains.
  • The persistent near-zero accuracy on Water Accumulation, Soil Liquefaction, and ASR provides a concrete, class-targeted data-augmentation agenda for subsequent work.
  • Because the pipeline is parameterized by only the document corpus and the label taxonomy, the same recipe can be re-instantiated for tunnel, pavement, mechanical, or medical diagnosis.

Where Pith is reading between the lines

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

  • The paper's 13-point generalization advantage is its most consequential claim but also its most fragile; an audit that releases the 100 inputs and has independent inspectors re-label them could easily overturn the 'quantization as regularizer' story if the set is non-representative.
  • Because the same manual PDFs seed both the retrieval store and the training data, test descriptions paraphrased from those manuals may retrieve their own block's triples, artificially inflating accuracy; a cleaner test would compute held-out manual blocks and query retrieval only from the remaining manuals.
  • The inference pipeline still requires an external large language model to judge triple relevance, which adds latency and an extra 5–10 GB-class memory footprint that partially offsets the encoder's savings; replacing the LLM filter with a small learned relevance classifier is an obvious next step the paper mentions but does not explore.
  • The paper's imbalance-ratio guideline (use 4-bit quantization when IR < 10, avoid it when IR > 20) is derived from three data points; extrapolating to other datasets would require controlled experiments varying IR and quantization levels independently.

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

4 major / 8 minor

Summary. The paper proposes a retrieval-augmented fine-tuning pipeline for classifying the cause of bridge damage from Japanese text descriptions. In Phase A, causal triples (subject, relation, object) are extracted from 15–35 diagnostic PDF manuals via an LLM and indexed in a FAISS vector store. In Phase B, for each labeled damage block, top-k triples are retrieved and filtered by an LLM relevance judge; the object texts are appended to the input. Phase C fine-tunes BERT-Large-Japanese with LoRA, QLoRA, and a QA-LoRA approximation; Phase D applies the same retrieval augmentation at inference. The central claim is that QLoRA matches full-precision LoRA on a 116-sample Golden Testset (87.07% each) while using 72% less GPU memory and 11% lower inference latency, and that QLoRA generalizes better on a 100-sample diverse evaluation (47.0% vs. 34.0%). The paper also proposes a difficulty-tagged Golden Testset and analyzes class-wise failure modes.

Significance. If the stated results hold, the paper would provide a practical recipe for memory-efficient fine-tuning of encoder models for domain-specific diagnostic classification, with a reusable evaluation benchmark and an interesting empirical observation about quantization acting as a regularizer. The four-phase pipeline is potentially transferable to other inspection or diagnostic domains. However, the significance is currently conditional: the main comparative claims rest on a single 116-sample testset, an unreleased author-constructed 100-sample set, and a retrieval pipeline that appears to leak test-block context into the augmented inputs. These issues make the headline accuracy numbers unreliable as evidence about fine-tuning strategies, though the methodology and benchmark construction are worthy of further study.

major comments (4)
  1. [§4.1, §4.4, §5.6] Retrieval leakage likely contaminates the Golden Testset. Phase A builds the FAISS index from all 35 PDFs before any train/test split, and the Golden Testset (Section 5.6) is a stratified split of the merged v0.4 corpus with no exclusion of test source blocks from the triple store. At inference (Section 4.4), each test description passes through the same retrieval pipeline; since a test block's own extracted triples are in the index, the top-k candidates will include them and the LLM relevance filter will retain them. The object text o_j in Eq. (3) then names the causal mechanism, effectively supplying the label in the augmented input. The reported 87.07% accuracy for both LoRA and QLoRA may therefore measure retrieval of the correct answer rather than encoder quality or fine-tuning differences. This also threatens the 100-sample evaluation in Section 5.8 if that set is drawn from or pro
  2. [§4.3.4, Table 2] The comparative claims are not supported by statistical evidence. All results appear to come from single-seed runs with no confidence intervals or significance tests. The Golden Testset has 116 samples, so a difference of 1.73 percentage points (QA-LoRA vs. LoRA) is well within sampling variability, and the exact equality of 87.07% for LoRA and QLoRA is not a meaningful ‘match’ under this precision. The 13-point gap on the 100-sample diverse evaluation in Section 5.8 also lacks any uncertainty quantification. Please provide multiple seeds, bootstrap confidence intervals, or a significance test, and temper claims accordingly.
  3. [§5.8] The 100-sample diverse evaluation is the main evidence for the recommended QLoRA strategy, but the set is unreleased and its construction protocol is not described beyond ‘10 per class’ and ‘different damage locations, mechanisms, and structural elements’. There is no specification of how the inputs were sampled, whether they overlap with the training corpus, or whether retrieval from the same FAISS index is used. Without the data or a detailed protocol, the 47.0% vs. 34.0% generalization gap cannot be independently verified or interpreted. Please release the set (or a substantial sample) and document the selection process, including how it relates to the Golden Testset and the triple index.
  4. [§4.3.3, Table 2] The method labeled ‘QA-LoRA’ is not QA-LoRA. Section 4.3.3 explicitly defines a ‘practical approximation’ that adds L2 regularization to LoRA adapters without the LoftQ initialization central to QA-LoRA [6]. Yet Table 2, the abstract, and Section 4.3.4 present this variant as ‘QA-LoRA’ and draw conclusions about its comparative performance. This is a mislabeling of the method under test. Rename the variant (e.g., ‘LoRA + L2’) and, if actual QA-LoRA is to be evaluated, implement the original method or clearly state that the comparison is with an approximation.
minor comments (8)
  1. [Abstract/§5.6] The Golden Testset is claimed as a ‘reusable benchmark contribution,’ but no URL, repository, or data-availability statement is provided. Please include a release mechanism.
  2. [§5.6] The phrase ‘stratified, deduplicated, and difficulty-tagged’ is useful, but the difficulty tagging criteria are informal (‘text length, number of retrieved triples, training stage, membership in rare classes’). Define the exact rule or at least provide the distribution of Easy/Medium/Hard per class.
  3. [Table 2] The table reports ‘F1 w’ but the text does not define whether it is the weighted F1 over all classes or the macro-averaged version. Please clarify and add class-wise F1 if available.
  4. [§5.3, Table 4] The row labels ‘LoRA Golden (536 train)’, ‘QLoRA Golden (536 train)’, and ‘QA-LoRA Golden (536 train)’ are clear, but the table lacks the validation accuracy and standard deviation. Consider adding these for consistency with the learning-curve discussion.
  5. [§6.2, Table 6] The IR-based guideline (IR<10: use QLoRA; IR>20: no quantization) is derived from three data points (v0.2, v0.3, v0.4). The paper honestly flags this as heuristic, but the text should also note that the trend line in Figure 7 is a hand-drawn interpolation, not a fitted model.
  6. [Figures 1 and 2] The pipeline figures are dense and the text in the subfigures is small. Please provide a vector-resolution version and use consistent terminology (e.g., ‘C2 prediction’ vs. ‘cause label’).
  7. [§4.1] The OCR pipeline reports ‘870 text blocks and 851 figure/table blocks’ for 35 PDFs, but Figure 1 lists ‘872 / 1721’ for the same phase. Please reconcile these numbers.
  8. [References] Reference [12] is cited as ‘arXiv preprint, 2026’ with no identifier; please complete the citation or remove it.

Circularity Check

1 steps flagged

Golden Testset accuracy is partly circular: test blocks come from the same corpus that populates the inference-time FAISS index, so Eq. (3) can place the block's own causal-explanation object into the input.

specific steps
  1. self definitional [Sections 4.1, 4.2, 4.4, 5.6; Eqs. (2)-(3)]
    "The augmented input to the encoder is: xi = [CLS] Si [SEP] o1 · o2 · · · · · o|Ci| [SEP] (3) ... The full knowledge index is T = ∪_{b∈B} E(b) (Phase A) ... we construct a Golden Testset using a stratified, deduplicated split of the merged v0.4 corpus (767 samples total) (Sec. 5.6) ... At inference, new damage text S_test^i undergoes the same retrieval pipeline (FAISS top-k + LLM YES/NO filter) to construct context ^Ci (Sec. 4.4)."

    Phase A builds the FAISS triple index from every block of the 35-PDF corpus before any train/test split, and Section 5.6 draws the test split from that same merged corpus. Deduplication removes duplicate samples but is not described as removing a test block's own triples from T. Since Eq. (3) defines the encoder input as Si plus the retrieved object texts, and Phase D reuses the same index, a test block can retrieve the triple extracted from its own source block; the object text of that triple names the causal mechanism, effectively supplying the label. The reported 87.07% accuracy for both LoRA and QLoRA on the Golden Testset is therefore not a clean prediction of the fine-tuning method—for leaked samples, the answer is already present in the input by construction.

full rationale

This paper is an empirical benchmark comparison rather than a derivation from fitted constants, so most circularity patterns (fitted-parameter-called-prediction, imported uniqueness theorem, ansatz-by-citation) do not apply. The author self-citation [12] appears only in related work and does not support the central claim. The one load-bearing circularity is in the Golden Testset evaluation: the retrieval index is built from the same corpus from which the test split is drawn, with no described exclusion of test blocks from the index, so the test-time retrieval context can contain triples generated from the very text that defines the ground-truth label. This makes the equal-accuracy result between LoRA and QLoRA at least partially an artifact of the retrieval store rather than a clean comparison of the encoders. The 100-sample diverse evaluation and class-wise failure analysis are author-constructed and unreleased, which is a reproducibility and selection-risk concern but not a circularity. The speed and GPU-memory measurements are independent empirical facts. Overall, the central fine-tuning comparison is partially circular, while other reported results retain independent content.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The central comparison rests on the quality of LLM-extracted triples and the fairness of the author-constructed test and evaluation sets; neither is externally verified. Hyperparameters and thresholds are hand-set, and no new physical or model entities are introduced.

free parameters (5)
  • retrieval threshold delta = 0.5
    Minimum cosine similarity for FAISS retrieval (Section 4.2); hand-set, determines which triples become context.
  • top-k triples = 10
    Number of retrieved triples per query before LLM filtering (Section 4.2); hand-set.
  • balanced sampling cap n_max = 100
    Per-class training-sample cap in Equation 8; hand-set, reshapes class distribution and affects all downstream accuracy numbers.
  • QA-LoRA L2 coefficient lambda = 0.01
    Regularization weight in Equation 11; authors admit smaller lambda may change QA-LoRA results (Section 4.3.4).
  • LoRA rank / alpha / dropout = 16 / 32 / 0.1
    LoRA hyperparameters chosen without systematic search; central to the fine-tuning comparison.
axioms (4)
  • domain assumption Qwen2.5-7B extracts accurate causal triples from OCR'd manual blocks, and the LLM YES/NO filter reliably removes irrelevant triples
    The entire training and inference context C_i (Equations 2 and 6) depends on LLM extraction and filtering; no human evaluation or consistency check is reported (Sections 4.1-4.2).
  • domain assumption The 35 Japanese diagnostic manuals and the 10-class taxonomy C2 cover the real damage-cause distribution for bridge inspection
    Labels and triples are derived from this corpus with no external ground-truth validation (Sections 3.3 and 5.2).
  • domain assumption Visible damage descriptions S_i and latent causes are linguistically separable enough for BERT to classify
    Stated as a sufficient condition for transfer (Section 3.4) and assumed throughout the classification task.
  • ad hoc to paper 4-bit quantization noise acts as implicit regularization that improves out-of-distribution accuracy
    This is a post-hoc explanation for the 13-point generalization gap (Section 6.1), not an independently tested mechanism.

pith-pipeline@v1.3.0-alltime-deepseek · 12753 in / 11891 out tokens · 119447 ms · 2026-08-01T07:51:51.002173+00:00 · methodology

0 comments
read the original abstract

Bridge infrastructure deteriorates gradually, yet its root causes---salt intrusion, freezing, fatigue cracking, and others---remain invisible to the naked eye. Expert diagnosis relies on tacit knowledge built over years of practice. We address the challenge of automating this latent causal reasoning by proposing a Damage Cause Encoder that classifies 10-class damage causes from visible damage descriptions $S_i$ for use in autonomous bridge diagnostic agents. Our approach chains three components: (i)Knowledge Triple Extraction---a large language model extracts causal triples of the form (damage $\xrightarrow{\mathtt{caused\_by}}$ cause) from 15--35 diagnostic PDF manuals and indexes them in a FAISS vector store; (ii)Retrieval-Augmented Context---at training and inference time, relevant causal triples $\mathcal{C}_i$ are retrieved and concatenated with $S_i$, converting implicit domain knowledge into explicit Encoder context; (iii)Systematic Fine-tuning Comparison---we conduct a rigorous comparison of LoRA, QLoRA, and QA-LoRA on a fixed Golden Testset (116 stratified samples), demonstrating that QLoRA achieves the optimal trade-off: identical test accuracy (87.07%) to full-precision LoRA, 11% faster inference, 72% lower GPU memory, and superior generalization across diverse unseen inputs. A controlled Golden Testset---stratified, deduplicated, and difficulty-tagged---is introduced as a reusable benchmark contribution. QLoRA further outperforms LoRA by 13 percentage points on a 100-sample diverse evaluation spanning all 10 damage cause classes.These findings enable memory-efficient, high-accuracy diagnostic agents on consumer-grade hardware for edge deployment.

Figures

Figures reproduced from arXiv: 2607.21680 by Takato Yasuno.

Figure 1
Figure 1. Figure 1: Pipeline Phases A–B. Phase A: PDF manuals are processed by two OCR engines (pypdfium2 for text, PaddleOCR for figures/tables) to extract blocks; Qwen2.5-7B then extracts causal triples (σj , rj , oj ) indexed in a 1024-dimensional FAISS store. Phase B: For each labeled block (Si , ci), the FAISS index is queried to retrieve top-k triples, filtered by an LLM relevance judge, forming context Ci . Balanced sa… view at source ↗
Figure 2
Figure 2. Figure 2: Pipeline Phases C–D. Phase C: BERT-Large-Ja (344M params) is fine-tuned with QLoRA (rec￾ommended); the highlighted FP16 Classifier Replacement resolves the BitsAndBytes incompatibility (Section 4.3.2), followed by 4-bit NF4 quantization, LoRA adapter injection (r=16, α=32), and weighted cross-entropy training. Phase D: At inference, the same retrieval pipeline augments a new damage descrip￾tion Si with con… view at source ↗
Figure 3
Figure 3. Figure 3: Class distribution comparison among v0.2 [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Confusion matrices for LoRA, QLoRA, and QA-LoRA on the Golden Testset (116 samples). [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Class-wise accuracy on 100 diverse samples (10 per class). Dashed lines indicate overall accuracy: [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Accuracy gain of QLoRA over LoRA as a function of Imbalance Ratio. The green region (IR < 12) indicates where 4-bit quantization acts as a regularizer and improves accuracy; the red region (IR > 12) indicates degradation. 6.4 Limitations Triple extraction quality. Qwen2.5-7B occasion￾ally produces malformed JSON or semantically im￾precise triples, requiring post-hoc filtering. The 10.6% of blocks producing… 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

26 extracted references · 5 linked inside Pith

  1. [1]

    Bridge and tunnel inspection status (in japanese).https://www.mlit.go.jp/ road/road/traffic/hyoushiki/index.html,

    Ministry of Land, Infrastructure, Transport and Tourism, Japan. Bridge and tunnel inspection status (in japanese).https://www.mlit.go.jp/ road/road/traffic/hyoushiki/index.html,

  2. [2]

    BERT: Pre-training of deep bidirectional transformers for language under- standing

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language under- standing. InProceedings of NAACL-HLT, pages 4171–4186. Association for Computational Lin- guistics, 2019

    2019

  3. [3]

    Japanese BERT- large (cl-tohoku/bert-large-japanese-v2)

    Tohoku NLP Group. Japanese BERT- large (cl-tohoku/bert-large-japanese-v2). https://huggingface.co/cl-tohoku/ bert-large-japanese-v2, 2023. Accessed: 2026

    2023

  4. [4]

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

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. InInterna- tional Conference on Learning Representations (ICLR), 2022

    2022

  5. [5]

    QLoRA: Efficient fine- tuning of quantized LLMs

    Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. QLoRA: Efficient fine- tuning of quantized LLMs. InAdvances in Neu- ral Information Processing Systems (NeurIPS), 2023

    2023

  6. [6]

    QA- LoRA: Quantization-aware low-rank adaptation of large language models.arXiv preprint arXiv:2309.14717, 2023

    Yuhui Xu, Lingxi Xie, Xiaotao Gu, Xiangnan Chen, Heng Chang, Henghui Zhang, Zhengsu Chen, Xiaopeng Zhang, and Qi Tian. QA- LoRA: Quantization-aware low-rank adaptation of large language models.arXiv preprint arXiv:2309.14717, 2023. 12

    Pith/arXiv arXiv 2023

  7. [7]

    Retrieval-augmented gen- eration for knowledge-intensive NLP tasks

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, et al. Retrieval-augmented gen- eration for knowledge-intensive NLP tasks. In Advances in Neural Information Processing Sys- tems (NeurIPS), 2020

    2020

  8. [8]

    Dense passage retrieval for open-domain question answering

    Vladimir Karpukhin, Barlas O˘ guz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering. InProceed- ings of EMNLP, 2020

    2020

  9. [9]

    From local to global: A graph RAG approach to query-focused sum- marization.arXiv preprint arXiv:2404.16130, 2024

    Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Tru- itt, and Jonathan Larson. From local to global: A graph RAG approach to query-focused sum- marization.arXiv preprint arXiv:2404.16130, 2024

    Pith/arXiv arXiv 2024

  10. [10]

    Bridge crack detection and classification using a two-stage deep convolutional neural network.Automation in Construction, 150:104821, 2023

    Byunghyun Kim and Soojin Cho. Bridge crack detection and classification using a two-stage deep convolutional neural network.Automation in Construction, 150:104821, 2023

    2023

  11. [11]

    Automated visual inspec- tion of bridge surface cracks using deep learning

    Kang Chen, Georg Reichard, Abiola Akanmu, and Xiangyu Xu. Automated visual inspec- tion of bridge surface cracks using deep learning. Computer-Aided Civil and Infrastructure Engi- neering, 37(2):193–212, 2022

    2022

  12. [12]

    Quantized vision-language mod- els for damage assessment: A comparative study of LLaV A-1.5-7B quantization levels

    Takato Yasuno. Quantized vision-language mod- els for damage assessment: A comparative study of LLaV A-1.5-7B quantization levels. arXiv preprint, 2026

    2026

  13. [13]

    A survey on large language model based autonomous agents.Frontiers of Computer Sci- ence, 18(6):186345, 2024

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, et al. A survey on large language model based autonomous agents.Frontiers of Computer Sci- ence, 18(6):186345, 2024

    2024

  14. [14]

    The rise and potential of large language model based agents: A survey.arXiv preprint arXiv:2309.07864, 2023

    Zhiheng Xi, Wenxiang Chen, Xin Guo, Wei He, et al. The rise and potential of large language model based agents: A survey.arXiv preprint arXiv:2309.07864, 2023

    Pith/arXiv arXiv 2023

  15. [15]

    Hudson, Ehsan Aditi, et al

    Rishi Bommasani, Drew A. Hudson, Ehsan Aditi, et al. On the opportunities and risks of foundation models.arXiv preprint arXiv:2108.07258, 2021

    Pith/arXiv arXiv 2021

  16. [16]

    HuggingGPT: Solving AI tasks with Chat- GPT and its friends in HuggingFace

    Yongliang Shen, Kaitao Song, Xu Tan, Dong- sheng Li, Weiming Lu, and Yueting Zhuang. HuggingGPT: Solving AI tasks with Chat- GPT and its friends in HuggingFace. InAd- vances in Neural Information Processing Sys- tems (NeurIPS), 2023

    2023

  17. [17]

    O’Brien, Carrie J

    Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Inter- active simulacra of human behavior. InProceed- ings of the ACM Symposium on User Interface Software and Technology (UIST), 2023

    2023

  18. [18]

    Farrar and Keith Worden

    Charles R. Farrar and Keith Worden. An intro- duction to structural health monitoring.Philo- sophical Transactions of the Royal Society A, 365(1851):303–315, 2007

    2007

  19. [19]

    Rui Zhao, Ruqiang Yan, Zhenghua Chen, Kezhi Mao, Peng Wang, and Robert X. Gao. Deep learning and its applications to machine health monitoring.Mechanical Systems and Signal Pro- cessing, 115:213–237, 2019

    2019

  20. [20]

    Andrew K. S. Jardine, Daming Lin, and Dragan Banjevic. A review on machinery diagnostics and prognostics implementing condition-based main- tenance.Mechanical Systems and Signal Process- ing, 20(7):1483–1510, 2006

    2006

  21. [21]

    Deep learning-based crack damage detection using convolutional neural networks

    Young-Jin Cha, Wooram Choi, and Oral B¨ uy¨ uk¨ ozt¨ urk. Deep learning-based crack damage detection using convolutional neural networks. Computer-Aided Civil and Infrastructure Engi- neering, 32(5):361–378, 2017

    2017

  22. [22]

    Road damage detection and classification using deep neural networks with smartphone images

    Hiroya Maeda, Yoshihide Sekimoto, Toshikazu Seto, Takehiro Kashiyama, and Hiroshi Omata. Road damage detection and classification using deep neural networks with smartphone images. Computer-Aided Civil and Infrastructure Engi- neering, 33(12):1127–1141, 2018

    2018

  23. [23]

    Qwen2.5 technical report.arXiv preprint arXiv:2412.15115, 2025

    Qwen Team. Qwen2.5 technical report.arXiv preprint arXiv:2412.15115, 2025

    Pith/arXiv arXiv 2025

  24. [24]

    Static embedding japanese

    Hotchpotch. Static embedding japanese. https://huggingface.co/hotchpotch/ static-embedding-japanese, 2024. Accessed: 2026

    2024

  25. [25]

    LLM.int8(): 8-bit ma- trix multiplication for transformers at scale

    Tim Dettmers, Mike Lewis, Younes Belkada, and Luke Zettlemoyer. LLM.int8(): 8-bit ma- trix multiplication for transformers at scale. In Advances in Neural Information Processing Sys- tems (NeurIPS), 2022

    2022

  26. [26]

    PEFT: State-of-the-art parameter-efficient fine-tuning.https: //github.com/huggingface/peft, 2023

    Hugging Face. PEFT: State-of-the-art parameter-efficient fine-tuning.https: //github.com/huggingface/peft, 2023. Accessed: 2026. 13

    2023