pith. sign in

arxiv: 2606.29706 · v1 · pith:FCT65DVCnew · submitted 2026-06-29 · 💻 cs.IR · cs.AI· cs.CL· cs.LG

ARMOR: Adaptive Retriever Optimization for Low-Resource Telecom Question Answering

Heshan Fernando , Quan Xiao , Yan Xin , Tianyi Chen This is my paper

Pith reviewed 2026-06-30 04:50 UTC · model grok-4.3

classification 💻 cs.IR cs.AIcs.CLcs.LG
keywords telecom QAretrieval-augmented generationquery encoder adaptationRAG optimizationlow-resource domainsInfoNCEcontrastive objectivesARMOR
0
0 comments X

The pith

Adapting only the query encoder in RAG improves telecom QA performance without generator fine-tuning.

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

The paper establishes that in low-resource telecom question answering, where evidence is fragmented across standards and technical documents, query-side retriever adaptation can outperform generator fine-tuning. It motivates this through a capacity argument comparing estimation terms under bounded parameters, then introduces ARMOR to jointly optimize a latent-document RAG likelihood and an InfoNCE contrastive objective while regularizing the adapted encoder toward its frozen base. Experiments across telecom retrieval and generative QA benchmarks show gains in evidence retrieval and answer generation when the generator remains fixed.

Core claim

ARMOR learns separate temperatures for the RAG retrieval distribution and the InfoNCE softmax, regularizes the adapted query encoder toward the frozen base, and targets downstream QA performance; under the paper's bounded-parameter and soft-retrieval assumptions this yields a smaller estimation term than supervised fine-tuning when the query encoder's effective dimension is smaller, producing measurable improvements on in-domain telecom benchmarks.

What carries the argument

ARMOR (Adaptive Regularized Mixture Optimization for Retrievers), which jointly optimizes latent-document RAG likelihood and InfoNCE contrastive loss with per-objective temperature learning and regularization to the base encoder.

If this is right

  • Query-encoder tuning reduces the estimation term relative to generator fine-tuning when effective dimension is smaller.
  • Joint RAG-likelihood and InfoNCE optimization improves both retrieval geometry and generation utility.
  • Regularization to the frozen base encoder limits over-specialization while retaining general capability.
  • Fixed-generator RAG systems achieve higher evidence retrieval and answer generation accuracy in several in-domain telecom settings.

Where Pith is reading between the lines

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

  • The same query-adaptation approach could be tested on other fragmented technical domains such as legal or medical QA without retraining the generator.
  • Keeping the generator frozen may preserve performance on out-of-domain or general questions while domain adaptation occurs only on the retriever side.
  • Varying the regularization strength or temperature schedules offers a direct experimental knob for balancing adaptation and stability.

Load-bearing premise

Query-encoder tuning produces a smaller estimation term than supervised fine-tuning when the effective dimension is smaller, under bounded-parameter and soft-retrieval assumptions.

What would settle it

A controlled experiment that applies generator fine-tuning at the same parameter budget as the query-encoder adaptation and measures whether retrieval or answer quality on the telecom benchmarks is equal or better.

Figures

Figures reproduced from arXiv: 2606.29706 by Heshan Fernando, Quan Xiao, Tianyi Chen, Yan Xin.

Figure 1
Figure 1. Figure 1: Comparison of optimization targets and the performance of retriever-training objec￾tives for ISAC domain QA with Llama-3-8B-Instruct [11] as the generator model and e5-large-v2 [38] as the dense retriever backbone. Retriever-side query-encoder optimization produces substan￾tially larger in-domain gains than either the base generator or generator-side adaptation, motivating our study of which component to o… view at source ↗
Figure 2
Figure 2. Figure 2: Training dynamics of adaptive temperatures and query-distillation regularization across ISAC, JCC, and SAGIN domains. Retrieval temperature consistently sharpens during training, while query-distillation loss rises late, indicating increasing tension between domain special￾ization and compatibility with the frozen embedding space. temperatures downward, eventually making the softmaxes excessively sharp. In… view at source ↗
Figure 3
Figure 3. Figure 3: ISAC Tele-Eval data and corpus ablations. (a) and (b) vary ARMOR training supervision and source-document coverage; (c) restricts training to one source category. Performance improves overall but remains sensitive to source alignment and corpus composition. REPLUG-style prediction can incur additional test-time overhead from document-wise scoring or aggregation, whereas our method retains the standard RAG-… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of Base Gen, Base RAG, and ARMOR across generator backbones on ISAC Tele-Eval. ARMOR’s gains are clearest for 8B-scale generators, suggesting that stronger generators are better able to use improved retrieved evidence [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
read the original abstract

Telecom question answering (QA) is a challenging setting for retrieval-augmented generation (RAG): evidence is fragmented across standards, papers, encyclopedic resources, and web documents, and answers often hinge on technical tables, equations, and specialized protocol language. In low-resource subdomains, generator fine-tuning can over-specialize and degrade general capability, making query-side retriever adaptation an attractive alternative. To this end, we ask whether a fixed-generator, query-adapted RAG system can outperform generator-side adaptation, and which retriever objectives best support that setting. We motivate retrieval, rather than generator fine-tuning, as the adaptation target through a capacity comparison: under bounded-parameter and soft-retrieval assumptions, query-encoder tuning can have a smaller estimation term than supervised fine-tuning when its effective dimension is smaller. We identify two particularly relevant objectives -- the latent-document RAG likelihood, which optimizes generation utility, and the InfoNCE contrastive objective, which improves semantic retrieval geometry -- and leverage them jointly through a retriever optimization method targeting downstream QA performance in the telecom domain. Specifically, we introduce ARMOR, Adaptive Regularized Mixture Optimization for Retrievers, which learns separate temperatures for the RAG retrieval distribution and InfoNCE softmax and regularizes the adapted query encoder toward the frozen base query encoder. Across telecom-specific retrieval and generative QA benchmarks, we show that ARMOR improves evidence retrieval and answer generation in several in-domain settings. Code is available at https://github.com/heshandevaka/ARMOR.git.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes ARMOR, an adaptive retriever optimization method for low-resource telecom QA in RAG systems. It keeps the generator fixed and adapts only the query encoder by jointly optimizing a latent-document RAG likelihood objective and an InfoNCE contrastive objective, learning separate temperatures for each and regularizing the adapted encoder toward the base model. The approach is motivated by a capacity comparison claiming that query-encoder tuning yields a smaller estimation term than generator fine-tuning under bounded-parameter and soft-retrieval assumptions when effective dimension is smaller. Empirical results are reported to show gains in evidence retrieval and answer generation on telecom-specific benchmarks.

Significance. If the empirical gains hold under proper controls and the capacity argument is made rigorous, the work could support practical query-side adaptation strategies in specialized domains where generator fine-tuning risks capability degradation. Code release is a positive factor for reproducibility.

major comments (2)
  1. [Abstract / §1] Abstract and introduction: the central motivation rests on an un-derived capacity comparison stating that 'query-encoder tuning can have a smaller estimation term than supervised fine-tuning when its effective dimension is smaller.' No section defines the estimation term, specifies how effective dimension is computed for the query encoder versus the generator, or shows the dimension reduction holds under the telecom data regime. This assumption directly justifies the fixed-generator design; without the derivation the preference for retriever-side adaptation lacks theoretical support.
  2. [§3 / Experiments] §3 (method) and experimental sections: the joint objective combines RAG likelihood and InfoNCE with learned temperatures, but the manuscript does not report ablation results isolating the contribution of each temperature or the regularization term to the claimed QA gains. Without these controls it is unclear whether the reported improvements are attributable to the proposed ARMOR components or to generic contrastive tuning.
minor comments (1)
  1. Notation for the two temperatures (RAG retrieval distribution and InfoNCE softmax) should be introduced with explicit symbols and distinguished from any other temperature parameters in the base models.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight opportunities to strengthen the theoretical motivation and empirical analysis. We address each major comment below and commit to revisions that directly respond to the concerns raised.

read point-by-point responses

  1. Referee: [Abstract / §1] Abstract and introduction: the central motivation rests on an un-derived capacity comparison stating that 'query-encoder tuning can have a smaller estimation term than supervised fine-tuning when its effective dimension is smaller.' No section defines the estimation term, specifies how effective dimension is computed for the query encoder versus the generator, or shows the dimension reduction holds under the telecom data regime. This assumption directly justifies the fixed-generator design; without the derivation the preference for retriever-side adaptation lacks theoretical support.

    Authors: We agree that the capacity comparison is stated at a high level in the introduction without a full derivation. The manuscript motivates the comparison under bounded-parameter and soft-retrieval assumptions but does not define the estimation term or detail the effective-dimension calculation. In the revision we will add a formal derivation (new subsection in §2 plus appendix) that (i) defines the estimation term, (ii) specifies how effective dimension is obtained for the query encoder versus the generator, and (iii) discusses the conditions under which the dimension reduction holds in the low-resource telecom regime. This will supply the missing theoretical support for the fixed-generator design. revision: yes

  2. Referee: [§3 / Experiments] §3 (method) and experimental sections: the joint objective combines RAG likelihood and InfoNCE with learned temperatures, but the manuscript does not report ablation results isolating the contribution of each temperature or the regularization term to the claimed QA gains. Without these controls it is unclear whether the reported improvements are attributable to the proposed ARMOR components or to generic contrastive tuning.

    Authors: We concur that the current experiments do not isolate the learned temperatures or the regularization term. The joint objective is presented in §3, yet no component-wise ablations appear in §4. In the revised manuscript we will add a dedicated ablation table (and corresponding text) that systematically removes (a) the separate temperature parameters and (b) the regularization toward the base encoder, reporting retrieval and QA metrics for each variant. These controls will clarify the contribution of the ARMOR-specific elements beyond generic contrastive tuning. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's core claims rest on empirical results for ARMOR on telecom retrieval and QA benchmarks. The capacity comparison is introduced as a stated motivation under explicit assumptions (bounded-parameter, soft-retrieval, effective dimension) without any derivation, equation, or reduction to prior fitted values shown in the text. No self-definitional steps, fitted inputs renamed as predictions, or load-bearing self-citations appear. The method (joint RAG likelihood + InfoNCE with temperature regularization) is defined directly and evaluated externally, making the derivation self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Ledger extracted from abstract only; full paper would likely add more parameters and assumptions.

free parameters (1)
  • separate temperatures for RAG retrieval distribution and InfoNCE softmax
    Learned per objective to balance the two losses.
axioms (1)
  • domain assumption bounded-parameter and soft-retrieval assumptions imply smaller estimation term for query-encoder tuning than for generator fine-tuning
    Invoked to motivate focusing adaptation on the retriever.

pith-pipeline@v0.9.1-grok · 5820 in / 1104 out tokens · 62914 ms · 2026-06-30T04:50:33.483327+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

40 extracted references · 21 canonical work pages · 13 internal anchors

  1. [1]

    Fine-grained analysis of op- timization and generalization for overparameterized two-layer neural networks

    Sanjeev Arora, Simon Du, Wei Hu, Zhiyuan Li, and Ruosong Wang. Fine-grained analysis of op- timization and generalization for overparameterized two-layer neural networks. InInternational 11 conference on machine learning, pages 322–332, 2019

    2019

  2. [2]

    In-context retrieval-augmented language models.TACL, 2023

    Akari Asai et al. In-context retrieval-augmented language models.TACL, 2023

    2023

  3. [3]

    A statistical framework for data- dependent retrieval-augmented models

    Soumya Basu, Ankit Singh Rawat, and Manzil Zaheer. A statistical framework for data- dependent retrieval-augmented models. InInternational Conference on Machine Learning, pages 3197–3223, 2024

    2024

  4. [4]

    Improving language models by retrieving from trillions of tokens

    Sebastian Borgeaud et al. Improving language models by retrieving from trillions of tokens. ICML, 2022

    2022

  5. [5]

    Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks

    Zhao Chen et al. Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks. InICML, 2018

    2018

  6. [6]

    SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training

    Tianzhe Chu, Yuexiang Zhai, Jihan Yang, Shengbang Tong, Saining Xie, Dale Schuurmans, Quoc V . Le, Sergey Levine, and Yi Ma. Sft memorizes, rl generalizes: A comparative study of foundation model post-training.arXiv preprint arXiv:2501.17161, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  7. [7]

    Ragas: Automated Evaluation of Retrieval Augmented Generation

    Shahul Es et al. Ragas: Automated evaluation of retrieval-augmented generation.arXiv preprint arXiv:2309.15217, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  8. [8]

    Mitigating gradient bias in multi-objective learning: A provably convergent stochastic approach

    Heshan Fernando, Han Shen, Miao Liu, Subhajit Chaudhury, Keerthiram Murugesan, and Tianyi Chen. Mitigating gradient bias in multi-objective learning: A provably convergent stochastic approach. InInternational Conference on Learning Representations, 2023

    2023

  9. [9]

    Variance reduction can improve trade-off in multi-objective learning

    Heshan Fernando, Lisha Chen, Songtao Lu, Pin-Yu Chen, Miao Liu, Subhajit Chaudhury, Keerthiram Murugesan, Gaowen Liu, Meng Wang, and Tianyi Chen. Variance reduction can improve trade-off in multi-objective learning. InICASSP 2024–2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 6975–6979. IEEE, 2024

    2024

  10. [10]

    Understanding forgetting in llm supervised fine-tuning and preference learning–a convex optimization perspective.arXiv preprint arXiv:2410.15483, 2024

    Heshan Fernando, Han Shen, Parikshit Ram, Yi Zhou, Horst Samulowitz, Nathalie Baracaldo, and Tianyi Chen. Understanding forgetting in llm supervised fine-tuning and preference learning–a convex optimization perspective.arXiv preprint arXiv:2410.15483, 2024

    work page arXiv 2024

  11. [11]

    The Llama 3 Herd of Models

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models.arXiv preprint arXiv:2407.21783, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  12. [12]

    Realm: Retrieval-augmented language model pre-training

    Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Ming-Wei Chang. Realm: Retrieval-augmented language model pre-training. InICML, 2020

    2020

  13. [13]

    ORPO: Monolithic Preference Optimization without Reference Model

    Jiwoo Hong, Noah Lee, and James Thorne. Orpo: Monolithic preference optimization without reference model.arXiv preprint arXiv:2403.07691, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  14. [14]

    LoRA: Low-Rank Adaptation of Large Language Models

    Edward Hu et al. Lora: Low-rank adaptation of large language models.arXiv preprint arXiv:2106.09685, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  15. [15]

    Intuitive fine-tuning: Towards unifying sft and rlhf into a single process.arXiv preprint arXiv:2405.11870, 2024

    Ermo Hua, Biqing Qi, Kaiyan Zhang, Yue Yu, Ning Ding, Xingtai Lv, Kai Tian, and Bowen Zhou. Intuitive fine-tuning: Towards unifying sft and rlhf into a single process.arXiv preprint arXiv:2405.11870, 2024

    work page arXiv 2024

  16. [16]

    Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering

    Gautier Izacard and Edouard Grave. Leveraging passage retrieval with generative models for open domain question answering.arXiv preprint arXiv:2007.01282, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2007

  17. [17]

    Unsupervised Dense Information Retrieval with Contrastive Learning

    Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. Unsupervised dense information retrieval with contrastive learning. arXiv preprint arXiv:2112.09118, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  18. [18]

    Atlas: Few-shot Learning with Retrieval Augmented Language Models

    Gautier Izacard et al. Atlas: Few-shot learning with retrieval augmented language models. arXiv preprint arXiv:2208.03299, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  19. [19]

    Dense passage retrieval for open-domain question answering

    Vladimir Karpukhin, Barlas Oguz, Sewon Min, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering. InEMNLP, 2020. 12

    2020

  20. [20]

    Multi-task learning using uncertainty to weigh losses for scene geometry and semantics

    Alex Kendall, Yarin Gal, and Roberto Cipolla. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. InCVPR, 2018

    2018

  21. [21]

    Overcoming catastrophic forgetting in neural networks.PNAS, 2017

    James Kirkpatrick et al. Overcoming catastrophic forgetting in neural networks.PNAS, 2017

    2017

  22. [22]

    Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks.arXiv preprint arXiv:2005.11401, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2005

  23. [23]

    Speciality vs generality: An empirical study on catastrophic forgetting in fine-tuning foundation models

    Yong Lin, Lu Tan, Hangyu Lin, Zeming Zheng, Renjie Pi, Han Zhao, Yuan Yao, et al. Speciality vs generality: An empirical study on catastrophic forgetting in fine-tuning foundation models. arXiv preprint arXiv:2309.06256, 2023

    work page arXiv 2023

  24. [24]

    Conflict-averse gradient descent for multi-task learning.Advances in Neural Information Processing Systems, 34:18878–18890, 2021

    Bo Liu, Xingchao Liu, Xiaojie Jin, Peter Stone, and Qiang Liu. Conflict-averse gradient descent for multi-task learning.Advances in Neural Information Processing Systems, 34:18878–18890, 2021

    2021

  25. [25]

    Uft: Unifying supervised and reinforce- ment fine-tuning.arXiv preprint arXiv:2505.16984, 2025

    Mingyang Liu, Gabriele Farina, and Asuman Ozdaglar. Uft: Unifying supervised and reinforce- ment fine-tuning.arXiv preprint arXiv:2505.16984, 2025

    work page arXiv 2025

  26. [26]

    Tele-llms: A series of specialized large language models for telecommunications.arXiv preprint arXiv:2409.05314, 2024

    Ali Maatouk, Kenny Chirino Ampudia, Rex Ying, and Leandros Tassiulas. Tele-llms: A series of specialized large language models for telecommunications.arXiv preprint arXiv:2409.05314, 2024

    work page arXiv 2024

  27. [27]

    Teleqna: A benchmark dataset to assess large language models telecommunications knowledge.IEEE Network, 2025

    Ali Maatouk, Fadhel Ayed, Nicola Piovesan, Antonio De Domenico, Merouane Debbah, and Zhi- Quan Luo. Teleqna: A benchmark dataset to assess large language models telecommunications knowledge.IEEE Network, 2025

    2025

  28. [28]

    MIT press, 2018

    Mehryar Mohri, Afshin Rostamizadeh, and Ameet Talwalkar.Foundations of machine learning. MIT press, 2018

    2018

  29. [29]

    Generalization bound for a shallow transformer trained using gradient descent.Transactions on Machine Learning Research, 2026

    Brian Mwigo and Anirban Dasgupta. Generalization bound for a shallow transformer trained using gradient descent.Transactions on Machine Learning Research, 2026. URL https: //openreview.net/forum?id=t3iUeMOT8Z

    2026

  30. [30]

    Kilt: A benchmark for knowledge-intensive language tasks

    Fabio Petroni et al. Kilt: A benchmark for knowledge-intensive language tasks. InNAACL, 2021

    2021

  31. [31]

    Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

    Xiangyu Qi, Yi Zeng, Tinghao Xie, Ruoxi Jia, Prateek Mittal, and Peter Henderson. Fine-tuning aligned language models compromises safety, even when users do not intend to!arXiv preprint arXiv:2310.03693, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  32. [32]

    Multi-task learning as multi-objective optimization

    Ozan Sener and Vladlen Koltun. Multi-task learning as multi-objective optimization. In NeurIPS, 2018

    2018

  33. [33]

    On penalty-based bilevel gradient descent method

    Han Shen, Quan Xiao, and Tianyi Chen. On penalty-based bilevel gradient descent method. Mathematical Programming, 214(1–2):539–589, 2025. doi: 10.1007/s10107-025-02194-4

    work page doi:10.1007/s10107-025-02194-4 2025

  34. [34]

    Replug: Retrieval-augmented black-box language models

    Weijia Shi, Sewon Min, Michihiro Yasunaga, Minjoon Seo, Richard James, Mike Lewis, Luke Zettlemoyer, and Wen-tau Yih. Replug: Retrieval-augmented black-box language models. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 8371–8384, 2024. doi: 10.18653/...

    work page doi:10.18653/v1/2024.naacl-long.463 2024

  35. [35]

    Qwen3 Technical Report

    Qwen Team. Qwen3 technical report, 2025. URLhttps://arxiv.org/abs/2505.09388

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  36. [36]

    Sequence length independent norm-based generalization bounds for transformers

    Jacob Trauger and Ambuj Tewari. Sequence length independent norm-based generalization bounds for transformers. InInternational Conference on Artificial Intelligence and Statistics, pages 1405–1413, 2024

    2024

  37. [37]

    Representation Learning with Contrastive Predictive Coding

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding.arXiv preprint arXiv:1807.03748, 2018. 13

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  38. [38]

    Text Embeddings by Weakly-Supervised Contrastive Pre-training

    Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Rangan Majumder, and Furu Wei. Text embeddings by weakly-supervised contrastive pre-training. arXiv preprint arXiv:2212.03533, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  39. [39]

    Gradient surgery for multi-task learning

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. Gradient surgery for multi-task learning. InAdvances in Neural Information Processing Systems, 2020

    2020

  40. [40]

    keep": true/false,

    Tianjun Zhang, Shishir G. Patil, Naman Jain, Sheng Shen, Matei Zaharia, Ion Stoica, and Joseph E. Gonzalez. Raft: Adapting language models to domain-specific rag.arXiv preprint arXiv:2403.10131, 2024. 14 A Experiment Details In this section, we provide additional details on the data generation, baseline implementation, and experiment setup used in this pa...

    work page arXiv 2024