Pith. sign in

REVIEW 4 major objections 6 minor 33 references

A Driver Advisory System Based on Large Language Model for High-speed Train

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Domain fine-tuning and retrieval-augmented generation let a 6B-parameter Chinese LLM answer high-speed railway knowledge questions and produce train-class-specific fault-handling suggestions.

desk verdict A solid pipeline paper whose headline accuracy claim is undercut by an evaluation set built with the same model family that generated its training data. read the letter →

arxiv 2501.07837 v1 pith:AWJ3PM2D submitted 2025-01-14 cs.AI

Y.C. Luo , J. Xun , W. Wang , R.Z. Zhang , Z.C. Zhao This is my paper · ORCID
classification cs.AI
keywords high-speedtraindriveradvisorysystemlargelanguagemodelretrieval-augmentedgenerationdomainfine-tuningLoRAfaulthandlingrailwayknowledgequestionanswering
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that a comparatively small Chinese language model, fine-tuned on railway-specific material and augmented with retrieval, can act as an intelligent driver advisory system for high-speed trains. The authors report that domain fine-tuning improves answer accuracy by about 10% on average, letting the tuned model outperform several mainstream Chinese chat models on railway knowledge questions. Adding retrieval-augmented generation raises recall-oriented ROUGE scores by about 4% on average, with the biggest gains on railway regulations. Simulated traction-loss and sensor-fault cases show the system pulling the relevant emergency manual and giving train-class-specific handling suggestions. The practical stake is that a driver facing a fault would no longer have to wait for the onboard mechanic to decide the next action.

What carries the argument

The load-bearing object is the IDAS-LLM pipeline, whose three parts carry the argument: the Railway Training Dataset (RTD), 10,100 structured Q&A pairs generated by Qwen-14B-Chat from 776,000 tokens of exam-syllabus source text; Low-Rank Adaptation (LoRA), which fine-tunes only about 2 million of the 6 billion parameters; and a retrieval-augmented generation stage that vectorizes railway text chunks with Bge-zh-v1.5 into a ChromaDB knowledge base, retrieves the top five chunks above a similarity threshold, and asks the fine-tuned model to revise its answer using that context. The mechanism's work is to inject domain vocabulary and procedures cheaply, then ground the model's output in retrievable manual text rather than bare parametric memory.

What would settle it

Have qualified high-speed-train driver examiners blindly grade the 300 assessment answers for factual correctness, then compare the human grades with the BLEU/ROUGE ordering; if human-graded accuracy does not improve after fine-tuning and RAG, the central claim fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a comparatively small Chinese model, ChatGLM3-6B, can be turned into a railway-domain advisor by LoRA fine-tuning on 10,100 LLM-generated question-and-answer pairs drawn from the driver examination syllabus. The fine-tuned model answers railway knowledge questions more accurately than several general-purpose Chinese chat models, with the largest per-category gains in railway expertise. Adding a retrieval-augmented generation stage, which embeds railway manuals with Bge-zh-v1.5, stores them in ChromaDB, and feeds the top five similar chunks into the model's context, further improves recall-oriented ROUGE scores, especially for railway regulations. Simulated traction-loss and sensor-fault scenarios show the system retrieving the emergency manual for the correct train class and producing handling suggestions in the driver's language.

Load-bearing premise

The results are scored against reference answers that the same AI model family used to create the training data wrote by rephrasing exam questions, and no human railway expert checks those answers; if those references are wrong or stylistically biased, the measured gains may not mean real railway expertise.

Editorial extensions

If this is right

  • A 6-billion-parameter model with LoRA fine-tuning is enough for competitive railway-domain answering, so the framework does not depend on access to a frontier-scale model.
  • RAG raises the recall metrics on railway regulation questions by 9–11 percentage points, indicating that manual-based retrieval can materially ground regulatory answers.
  • The fault-case simulations show train-class-specific retrieval, so the same system can be extended to other rolling-stock families by adding their emergency manuals to the knowledge base.
  • The framework is presented as a practical tool for real operations, with the conclusion proposing future work on more realistic datasets, knowledge graphs, and real-time performance.

Reading between the lines

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

  • Editorial inference: because the reference answers were generated by the same model family that produced the training data, the biggest threat to the 10% and 4% numbers is reference bias; a human-scored blind evaluation on the same 300 questions would settle how much of the gain is real railway competence.
  • Editorial inference: the recipe, official exam syllabus, LLM-generated Q&A, LoRA fine-tuning, and manual-retrieval grounding, should transfer to other safety-regulated domains with standard exams and emergency handbooks, such as aviation, metro transit, or maritime operations.
  • Editorial inference: the chunk-size result (500 tokens best) suggests a testable extension where chunk boundaries are aligned to manual clauses or procedure steps rather than fixed token counts, which might improve retrieval precision further.
  • Editorial inference: for a safety-critical advisory role, lexical metrics alone cannot certify that a suggestion is safe; the framework would need a verification layer that refuses to answer when retrieved evidence is below threshold.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes IDAS-LLM, a driver advisory system for high-speed trains built by fine-tuning ChatGLM3-6B with LoRA on a generated railway Q&A dataset (RTD, 10,100 pairs) and augmenting it with retrieval-augmented generation (RAG) over a railway text database. It evaluates the fine-tuned model against several Chinese LLMs using BLEU and ROUGE on a 300-question assessment set, reports an average BLEU improvement of about 10 percentage points, and reports that RAG increases average ROUGE by about 4%. Two simulated fault-handling case studies for traction loss and sensor faults are presented as demonstrations of the framework's practical capability.

Significance. If the empirical claims were supported, the paper would provide a useful application recipe: it assembles a railway domain dataset grounded in the official examination syllabus, demonstrates that LoRA fine-tuning on a 6B model can be done on a single A40 GPU, and shows a concrete RAG pipeline with source citation for emergency manuals. The case studies illustrate a plausible human-system interaction for fault handling. The authors should be credited for grounding the dataset in a formal syllabus and for demonstrating the full pipeline on realistic fault scenarios. However, the paper does not release code or data, and the central evaluation currently rests on an internally generated reference set, so the significance of the numerical claims cannot be assessed until the evaluation is independently grounded.

major comments (4)
  1. [§3.1, Dataset for Evaluation; Tables 3–4] The reference answers used for scoring are produced by Qwen-14B-Chat, the same model family used to generate the 10,100 fine-tuning pairs in RTD. Since the fine-tuned ChatGLM3-6B is trained to imitate Qwen-style phrasing, its higher BLEU/ROUGE scores against Qwen-written references may reflect stylistic mimicry rather than railway expertise. The abstract's central claim of an 'improvement in answer accuracy by an average of 10%' is based on BLEU gains against these references, and the paper provides no human expert validation of either the reference answers or the model outputs. This dependency should be removed or explicitly controlled by using human-written reference answers, references produced by a different model family, or expert grading of factual correctness.
  2. [§4.1, Table 3; §4.2, Table 4] All quantitative claims rest on single-point BLEU/ROUGE scores from 100 questions per category, without confidence intervals, repeated sampling, or significance tests. Differences such as R1 0.47 vs 0.49 (legal provision, with RAG) and R2 0.16 vs 0.18 (railway expertise, with RAG) are within plausible sampling noise for n=100, so the reported improvements may not be reliable. The authors should report bootstrap confidence intervals or paired significance tests, and should describe how the 100 questions were sampled from the 2,462 collected.
  3. [§4.1, Fine-tuning Comparison Experiments] The text equates higher BLEU with higher accuracy and higher ROUGE with higher recall, but these metrics measure lexical n-gram overlap, not factual correctness. In a safety-critical domain such as train fault handling, the claim of improved 'answer accuracy' requires direct correctness evaluation, especially since the authors themselves note that BLEU/ROUGE scores are depressed by answer length and legitimate synonymy. Human expert scoring on a sample, or exact-answer matching for questions derived from multiple-choice and true/false items, should be reported alongside the automatic metrics.
  4. [§3.3, §4.2, Figure 3] The RAG evaluation reports only ROUGE changes against the same Qwen-generated references and never evaluates retrieval quality, such as whether the retrieved top-5 chunks actually contain the correct manual content. In addition, the chunk-size comparison in Figure 3 appears to be tuned on the same 100-question assessment set without error bars or a held-out split, so the 'about 4%' RAG recall improvement is not robustly established. Retrieval-side metrics (e.g., recall@k against a labeled query set) and an independent test set would make the claim testable.
minor comments (6)
  1. [Title] The title contains a typo: 'LAN GUAGE' should be 'LANGUAGE'.
  2. [Table 3] The Δ% row reports percentage-point differences but labels them as percentages; for example, BLEU 0.06→0.21 is a 15-percentage-point increase, not a 15% relative increase. The abstract's 'average of 10%' should use the same unambiguous units.
  3. [§4.2, Table 4] Table 4 omits BLEU without explanation, although the text says the same evaluation metrics as the fine-tuning experiments are used; clarify why BLEU is excluded from the RAG comparison.
  4. [Figure 3] Figure 3 does not label its axes or list the chunk sizes compared, and the line plots appear to lack error bars; the claim that 500-token chunks are best cannot be verified from the figure as presented.
  5. [§4.1 and References] The model name 'Internlm2-chat-7b' should be capitalized consistently, and 'CEval'/'C-Eval' should be unified.
  6. [§3.2, Supervised Fine-Tuning] The fine-tuning section reports 10 epochs and an 8:2 split but gives no validation-loss curve or early-stopping criterion; please report overfitting behavior for this relatively small dataset.

Circularity Check

2 steps flagged · score 6.0 of 10

The claimed 10% BLEU gain is measured against Qwen-14B-Chat-generated references while the fine-tuning data also came from Qwen-14B-Chat; the improvement partly reduces to stylistic imitation, and RAG chunk size is tuned on the evaluation set.

  1. fitted input called prediction [Methods, 'Dataset for Training' and 'Dataset for Evaluation'; Experiments, Table 3]
    "In conclusion, using the Qwen-14B-Chat (20) model on a single A40 GPU, 10,100 structured question-and-answer pairs are generated for fine-tuning. ... Similarly, by using the Qwen-14B-Chat model, single choice questions, multiple choice questions, and true/false questions from the question bank are transformed into a question-and-answer format to facilitate the evaluation of the framework's text generation performance. ... BLEU ... is a precision-based similarity measure that focuses on accuracy."

    The fine-tuning corpus and the evaluation references are both produced by Qwen-14B-Chat. SFT trains ChatGLM3-6B to maximize the likelihood of Qwen-generated answers, while BLEU/ROUGE score outputs against Qwen-generated references. A model that imitates Qwen's phrasing will therefore show higher BLEU/ROUGE by construction, independent of railway knowledge. The reported '10% average improvement in answer accuracy' is thus partly a measure of self-similarity to the reference generator, not externally validated domain expertise. The paper provides no human-expert check of either the references or the model's answers, and the conclusion's limitation paragraph mentions only residual hallucinations without acknowledging this dependency.

  2. other [Experiments, 'RAG comparison Experiments', Figure 3 and Table 4]
    "From the metrics in Figure 3, as they vary with chunk size, it is evident that chunking at 500 tokens yields better metrics across the three categories compared to other chunk sizes. This indicates that in the IDAS-LLM framework, a chunk size of 500 tokens is reasonable for textual data in the driving data knowledge base."

    The RAG chunk size (500 tokens) is selected by comparing evaluation metrics on the same assessment set, and the final RAG comparison in Table 4 is then reported on that same set. This is test-set tuning: the configuration is chosen to maximize the metric that later serves as evidence for the 'about 4%' recall improvement. The reported gain is therefore partly an artifact of selecting the hyperparameter on the evaluation data rather than an independent confirmation of the RAG design.

full rationale

The paper makes no use of self-citations for its central claims, and there are no equations that are circular by definition. However, the central empirical claim—that domain fine-tuning improves answer accuracy by an average of 10%—is evaluated with BLEU/ROUGE against reference answers generated by Qwen-14B-Chat, the same model used to generate the 10,100 fine-tuning pairs. Because SFT trains the base model to imitate Qwen-style outputs, the metric improvement is partially forced by construction: higher n-gram overlap with Qwen-generated references can reflect stylistic mimicry rather than railway expertise. The RAG 'recall' improvement is also reported after selecting the chunk size on the evaluation set, which inflates the apparent benefit. These issues compromise the validity of the headline numbers, but they do not make the entire derivation tautological: the fine-tuned model is a different architecture (ChatGLM3-6B), the training data and evaluation set are not identical, and the qualitative fault-handling case studies show retrieval of distinct manuals for different train classes. The paper would need human-expert validation or references generated without the training-data generator to support the accuracy claims independently. Overall, this is partial circularity in the evaluation, not a self-citation chain or a definitional identity, so a score of 6 is appropriate.

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

The central claim rests on three unverified assumptions: the Qwen-generated training and evaluation text is accurate, BLEU/ROUGE capture answer quality, and the sampled 100 questions per category are representative. Retrieval parameters such as chunk size are tuned on the test set, and the similarity threshold is unspecified.

free parameters (3)
  • RAG chunk size = 500 tokens
    Selected by comparing evaluation metrics across chunk sizes on the assessment set (Figure 3), so it is tuned to the test data.
  • Retrieved document count (top_k) = 5
    System retrieves the top five chunks by similarity with no sensitivity analysis reported.
  • RAG similarity threshold = not specified
    The paper says a preset threshold decides whether retrieved texts are relevant, but gives no value or tuning procedure.
assumptions (3)
  • domain assumption Qwen-14B-Chat-generated question-answer pairs faithfully represent railway knowledge and exam content.
    Both the RTD training data and the assessment reference answers are produced by Qwen-14B-Chat and filtered only by automated checks; no human expert verification is reported (Methods, Dataset for Training and Dataset for Evaluation).
  • domain assumption BLEU and ROUGE are valid proxies for answer accuracy and recall in railway fault-handling questions.
    The paper uses BLEU/ROUGE as the primary evaluation and interprets them as accuracy/recall while acknowledging their literal-overlap limitation (Experiments).
  • domain assumption The 100 sampled question-answer pairs per category are representative of the full railway driving knowledge assessment.
    The selection procedure for extracting 100 pairs per category from 2462 questions is not described, so representativeness is assumed (Dataset for Evaluation).

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Driver Advisory System Based on Large Language Model for High-speed Train." pith.science (2026). https://pith.science/paper/AWJ3PM2D

@misc{pith2026250107837,
  author       = {Pith},
  title        = {Pith review of: A Driver Advisory System Based on Large Language Model for High-speed Train},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AWJ3PM2D}},
  note         = {Machine review of arXiv:2501.07837}
}
read the original abstract

With the rapid development of China high-speed railway, drivers face increasingly significant technical challenges during operations, such as fault handling. Currently, drivers depend on the onboard mechanic when facing technical issues, for instance, traction loss or sensor faults. This dependency can hinder effective operation, even lead to accidents, while waiting for faults to be addressed. To enhance the accuracy and explainability of actions during fault handling, an Intelligent Driver Advisory System (IDAS) framework based on a large language model (LLM) named IDAS-LLM, is introduced. Initially, domain-fine-tuning of the LLM is performed using a constructed railway knowledge question-and-answer dataset to improve answer accuracy in railway-related questions. Subsequently, integration of the Retrieval-augmented Generation (RAG) architecture is pursued for system design to enhance the explainability of generated responses. Comparative experiments are conducted using the constructed railway driving knowledge assessment dataset. Results indicate that domain-fine-tuned LLMs show an improvement in answer accuracy by an average of 10%, outperforming some current mainstream LLMs. Additionally, the inclusion of the RAG framework increases the average recall rate of question-and-answer sessions by about 4%. Finally, the fault handling capability of IDAS-LLM is demonstrated through simulations of real operational scenarios, proving that the proposed framework has practical application prospects.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 23 canonical work pages

  1. [1]

    Bagheri, V

    Hani Tabai, B., M. Bagheri, V. Sadeghi-Firoozabadi, V. Shahidi, and H. Mirasadi, Impact of train drivers’ cognitive responses on rail accidents. Transportation research record, Vol. 2672, 2018, pp. 260–268

    work page 2018

  2. [2]

    Li, Z.-g

    Huang, Y.-c., L.-p. Li, Z.-g. Liu, H.-y. Zhu, and L. Zhu, Assessment of urban railway transit driver workload and fatigue under real working conditions. Transportation research record, Vol. 2673, 2019, pp. 891–900

    work page 2019

  3. [3]

    Zhu, Z.-g

    You, X.-d., L. Zhu, Z.-g. Liu, and Y.-c. Huang, Experimental study on the relationship between fatigue and unsafe behavior of urban rail transit drivers. Transportation research record, Vol. 2675, 2021, pp. 1151–1160

    work page 2021

  4. [4]

    Dong, and H

    Gao, S., H. Dong, and H. Zhu, Intelligent Driver Advisory System for Improving Manual Train Driving Performance. Journal of the China Railway Society, Vol. 41, 2019, pp. 88– 94

    work page 2019

  5. [5]

    Ouyang, L., J. Wu, X. Jiang, D. Almeida, C. Wainwright, P. Mishkin, C. Zhang, S. Agar- wal, K. Slama, A. Ray, et al., Training language models to follow instructions with human feedback. Advances in neural information processing systems, Vol. 35, 2022, pp. 27730 – 27744

    work page 2022

  6. [6]

    Cui, J., Z. Li, Y. Yan, B. Chen, and L. Yuan, Chatlaw: Open-source legal large language model with integrated external knowledge bases. arXiv preprint arXiv:2306.16092 , 2023

    arXiv 2023

  7. [7]

    Irsoy, S

    Wu, S., O. Irsoy, S. Lu, V. Dabravolski, M. Dredze, S. Gehrmann, P. Kambadur, D. Rosen- berg, and G. Mann, Bloomberggpt: A large language model for finance. arXiv preprint arXiv:2303.17564 , 2023

    arXiv 2023

  8. [8]

    Yang, S., H. Zhao, S. Zhu, G. Zhou, H. Xu, Y. Jia, and H. Zan, Zhongjing: Enhanc- ing the chinese medical capabilities of large language mo del through expert feedback and real-world multi -turn dialogue. In Proceedings of the AAAI Conference on Artificial Intel - ligence, 2024, Vol. 38, pp. 19368–19376

    work page 2024

Show all 33 references
  1. [9]

    Zhang, S., D. Fu, W. Liang, Z. Zhang, B. Yu, P. Cai, and B. Yao, Trafficgpt: Viewing, pro- cessing and interacting with traffic foundation models. Transport Policy, Vol. 150, 2024, pp. 95–105

    2024

  2. [10]

    Du, H., S. Teng, H. Chen, J. Ma, X. Wang, C. Gou, B. Li, S. Ma, Q. Miao, X. Na, et al., Chat with chatgpt on intelligent vehicles: An ieee tiv perspective. IEEE Transactions on Intelligent Vehicles, Vol. 8, 2023, pp. 2020–2026

    2023

  3. [11]

    Zhu, H., X. Sun, L. Chen, S. Gao, and H. Dong, Analysis and design of Driver Advi- sory System (DAS) for energy-efficient train operation with real-time information. In 2016 IEEE international conference on intelligent rail transportation (ICIRT), IEEE, 2016, pp. 99–104

    2016

  4. [12]

    Zhu, and S

    Dong, H., H. Zhu, and S. Gao, An approach for energy-efficient and punctual train op- eration via driver advisory system. IEEE Intelligent Transportation Systems Magazine, Vol. 10, 2018, pp. 57–67

    2018

  5. [13]

    Xiao, Z., Q. Wang, P. Sun, Z. Zhao, Y. Rao, and X. Feng, Real-time energy-efficient driver advisory system for high-speed trains. IEEE Transactions on Transportation Electrifica- tion, Vol. 7, 2021, pp. 3163–3172

    2021

  6. [14]

    Carbone, F

    Formato, G., L. Carbone, F. Grassi, R. Divano, and P. Segarich, Energy efficiency driv- ing strategy with Connected -Driving Advisory System (C-DAS). Transportation research procedia, Vol. 72, 2023, pp. 263–270. Y. C. Luo, J. Xun, W. Wang, R. Z. Zhang, Z. C. Zhao 17

    2023

  7. [15]

    Wang, P., R. M. Goverde, and J. Van Luipen, A connected driver advisory system frame- work for merging freight trains. Transportation Research Part C: Emerging Technologies, Vol. 105, 2019, pp. 203–221

    2019

  8. [16]

    Berdjag, P

    Guerra, T.-M., D. Berdjag, P. Polet, and T. A.-T. Nguyen, Toward a Cooperative ADAS for Train Driving based on Real-Time Human Parameters and Delay Estimation. IFAC- PapersOnLine, Vol. 56, 2023, pp. 3534–3539

    2023

  9. [17]

    Huang, J

    Cui, Y., S. Huang, J. Zhong, Z. Liu, Y. Wang, C. Sun, B. Li, X. Wang, and A. Khajepour, Drivellm: Charting the path toward fu ll autonomous driving with large language models. IEEE Transactions on Intelligent Vehicles, 2023

    2023

  10. [18]

    Cui, C., Y. Ma, X. Cao, W. Ye, and Z. Wang, Drive as you speak: Enabling human- like interaction wit h large language models in autonomous vehicles. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 902 –909

    2024

  11. [19]

    Wang, P., X. Wei, F. Hu, and W. Han, Transgpt: Multi-modal generative pre-trained trans- former for transportation. arXiv preprint arXiv:2402.07233 , 2024

    2024 arXiv

  12. [20]

    Bai, J., S. Bai, Y. Chu, Z. Cui, K. Dang, X. Deng, Y. Fan, W. Ge, Y. Han, F. Huang, B. Hui, L. Ji, M. Li, J. Lin, R. Lin, D. Liu, G. Liu, C. Lu, K. Lu, J. Ma, R. Men, X. Ren, X. Ren, C. Tan, S. Tan, J. Tu, P. Wang, S. Wang, W. Wang, S. Wu, B. Xu, J. Xu, A. Yang, H. Yang, J. Ya...

    2023 arXiv

  13. [21]

    Huang, Y., Y. Bai, Z. Zhu, J. Zhang, J. Zhang, T. Su, J. Liu, C. Lv, Y. Zhang, Y. Fu, et al., C-eval: A multi-level multi-discipline chinese evaluation suite for foundation models. Ad- vances in Neural Information Processing Systems, Vol. 36, 2024

    2024

  14. [22]

    Zhang, F

    Li, H., Y. Zhang, F. Koto, Y. Yang, H. Zhao, Y. Gong, N. Duan, and T. Baldwin, Cmmlu: Measuring massive multitask language understanding in chinese. arXiv preprint arXiv:2306.09212 , 2023

    2023 arXiv

  15. [23]

    Du, Z., Y. Qian, X. Liu, M. Ding, J. Qiu, Z. Yang, and J. Tang, GLM: General Language Model Pretraining with Autoregressive Blank Infilling. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2022, pp. 320–335

    2022

  16. [24]

    I., Baichuan -7B

    Technology, B. I., Baichuan -7B. https://github.com/baichuan -inc/Baichuan -7B? tab=readme-ov-file, 2023, accessed April 27, 2024

    2023

  17. [25]

    Muennighoff, N., T. Wang, L. Sutawika, A. Roberts, S. Biderman, T. L. Scao, M. S. Bari, S. Shen, Z. -X. Yong, H. Schoelkopf, et al., Crosslingual generalization through multitask finetuning. arXiv preprint arXiv:2211.01786 , 202 2

    arXiv

  18. [26]

    Hu, E. J., Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, and W. Chen, Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685 , 2021

    2021 arXiv

  19. [27]

    Xiao, S., Z. Liu, P. Zhang, and N. Muennighoff, C-Pack: Packaged Resources To Advance General Chinese Embedding, 2023

    2023

  20. [28]

    Roukos, T

    Papineni, K., S. Roukos, T. Ward, and W.-J. Zhu, Bleu: a method for automatic evaluation of machine translation. In Proceedi ngs of the 40th annual meeting of the Association for Computational Linguistics, 2002, pp. 311 –318

    2002

  21. [29]

    -Y., Rouge: A package for automatic evaluation of summaries

    Lin, C. -Y., Rouge: A package for automatic evaluation of summaries. In Text summariza- tion branches out, 2004, pp. 74–81. Y. C. Luo, J. Xun, W. Wang, R. Z. Zhang, Z. C. Zhao 18

    2004

  22. [30]

    Cai, Z., M. Cao, H. Chen, K. Chen, K. Chen, X. Chen, X. Chen, Z. Chen, Z. Chen, P. Chu, X. Dong, H. Duan, Q. Fan, Z. Fei, Y. Gao, J. Ge, C. Gu, Y. Gu, T. Gui, A. Guo, Q. Guo, C. He, Y. Hu, T. Huang, T. Jiang, P. Jiao, Z. Jin, Z. Lei, J. Li, J. Li, L. Li, S. Li, W. Li, Y. Li, H...

    2024

  23. [31]

    Yang, A., B. Yang, B. Hui, B. Zheng, B. Yu, C. Zhou, C. Li, C. Li, D. Liu, F. Huang, et al., Qwen2 technical report. arXiv preprint arXiv:2407.10671 , 2024

    2024 arXiv

  24. [32]

    Yang, A., B. Xiao, B. Wang, B. Zhang, C. Bian, C. Yin, C. Lv, D. Pan, D. Wang, D. Yan, et al., Baichuan 2: Open large-scale language models. arXiv preprint arXiv:2309.10305 , 2023

    2023 arXiv

  25. [33]

    Xiong, X

    Gao, Y., Y. Xiong, X. Gao, K. Jia, J. Pan, Y. Bi, Y. Dai, J. Sun, and H. Wang, Retrieval-augmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997 , 2023

    2023 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.