HANRAG: Heuristic Accurate Noise-resistant Retrieval-Augmented Generation for Multi-hop Question Answering

Dan Yang; Duolin Sun; Jian Wang; Jie Feng; Jinjie Gu; Lianzhen Zhong; Peng Wei; Yihan Jiao; Yue Shen; Zhehao Tan

REVIEW 4 major objections 7 minor 12 references

HANRAG: Heuristic Accurate Noise-resistant Retrieval-Augmented Generation for Multi-hop Question Answering

T0 review · 4 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read HANRAG claims that a single trained 'Revelator' agent can route, decompose, and noise-filter queries well enough to beat Adaptive-RAG on single-hop and multi-hop question answering while using fewer retrieval steps.

desk verdict Worth engaging with: a real efficiency idea and a sensible compound/complex distinction, but the single-hop results are not credible as printed and the fair comparison should lead. read the letter →

arxiv 2509.09713 v1 pith:CGVWIY2W submitted 2025-09-08 cs.CL cs.AI

Duolin Sun , Dan Yang , Yue Shen , Yihan Jiao , Zhehao Tan , Jie Feng , Lianzhen Zhong , Jian Wang

show 2 more authors

Peng Wei Jinjie Gu

This is my paper · ORCID

classification cs.CLcs.AI

keywords retrieval-augmentedgenerationmulti-hopquestionansweringqueryroutingdecompositionrelevancefilteringnoiseresistancecompoundqueriescomplex

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

The paper tries to establish that a single trained agent, the Revelator, can make retrieval-augmented question answering both more accurate and cheaper by deciding what kind of question it faces before retrieving. The key move is to split multi-hop questions into compound ones, whose sub-questions are independent and answerable in parallel, and complex ones, which require a dependent chain of reasoning answered step by step. A relevance filter then removes retrieved documents judged irrelevant, preventing noise from accumulating across iterations. The reported result is that HANRAG beats Adaptive-RAG and other baselines on EM, F1, and accuracy across single-hop, multi-hop, and a newly built compound benchmark, while using fewer retrieval steps. If the claim is right, a relatively small fine-tuned model can replace hand-tuned retrieval policies.

What carries the argument

The load-bearing object is the Revelator, a single fine-tuned 8B instruction-following language model that performs five judgments in one network: routing the query class, decomposing compound questions into sub-questions, refining the next seed question for complex questions, deciding whether a retrieved document is relevant, and deciding whether the iterative loop can end. The paper also reports a fair-comparison variant with a smaller 3B generator to show that the gains are not simply a bigger generator. The Revelator lets the framework replace a fixed iterative retrieval policy with a per-query pipeline, and its relevance judgment keeps noisy documents away from the generator.

What would settle it

Take the MuSiQue test set and inject, for each query, several retrieved passages that share named entities and topical words with the query but do not contain the answer, then compare HANRAG with and without the relevance filter. If accuracy falls to the no-filter ablation level (37.8 Acc) rather than holding near the full-system level (43.2 Acc), the claimed noise resistance does not generalize beyond the training distribution of the relevance discriminator.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the dominant source of error in multi-hop RAG is not the retriever but the absence of query-type awareness and post-retrieval noise control. HANRAG's Revelator first classifies each query into one of four routes, then acts as decomposer for compound questions, as refiner of seed questions for complex questions, and as a relevance discriminator after every retrieval. The compound-versus-complex distinction is presented as the conceptual fix for wasted iterative retrieval: compound sub-questions are retrieved asynchronously in one round, while complex seed questions are refined and retrieved synchronously until an ending discriminator halts the loop. Across three single-hop, three complex multi-hop, and one new compound benchmark, HANRAG is reported to outperform Adaptive-RAG on EM, F1, and accuracy, with the largest accuracy gain on the compound benchmark (71.76 vs 52.13) and with fewer average retrieval steps. The ablation on MuSiQue shows that the relevance discriminator and the refiner each carry a large part of the accuracy, while the ending discriminator is what keeps the step count low.

Load-bearing premise

The load-bearing premise is that the Revelator can recognize whether a retrieved document is relevant well enough to trust it on new queries, not just on the (query, document) pairs used to train it; the ablation shows that when that filter is removed, MuSiQue accuracy drops from 43.2 to 37.8, so the whole gain rests on this judgment generalizing.

Editorial extensions

If this is right

Compound multi-hop questions can be answered in close to one retrieval step instead of several, because sub-questions are retrieved in parallel and merged.
Post-retrieval relevance filtering improves single-hop accuracy even when the retriever is fixed: HANRAG's EM on SQuAD rises from Adaptive-RAG's 26.8 to 39.8.
Complex multi-hop questions gain accuracy from explicit seed-question refinement, and an ending discriminator prevents unnecessary retrieval-generation cycles.
On the newly constructed compound benchmark, the reported accuracy gain over Adaptive-RAG is 19.63 points with about 1.5 fewer retrieval steps on average.

Reading between the lines

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

The compound-versus-complex split is a cheap textual signal that other pipelines could adopt without the full Revelator, since a classifier alone might reproduce much of the step reduction.
Because the relevance labels and the compound benchmark are both generated by a single large language model, the framework's generalization to user-issued queries and to documents outside Wikipedia's style remains the main untested boundary.
A natural extension is to plug the same Revelator into a dense retriever or a hybrid retriever; if the gains persist, the framework is retriever-agnostic rather than tuned to BM25's noise profile.
The reported routing accuracy of 83.93% suggests that a targeted study of routing errors, especially straightforward questions sent to retrieval, could yield further gains without new retrieval logic.

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.

Desk Editor's Note

read the letter

First, the thing you need to know: this is a coherent systems paper with a genuinely useful efficiency story, but the headline numbers in Table 1 are not credible as printed, and the authors need to fix that before you trust the single-hop claims.

The new part is the Revelator, a single fine-tuned model that routes queries, decomposes compound questions, refines seed questions for complex ones, filters retrieved documents by relevance, and decides when to stop iterating. That combination is new, and the distinction between compound and complex multi-hop queries is worth taking seriously. The asynchronous retrieval for compound queries gives real step reductions—about half the retrieval steps on their compound benchmark. And importantly, the fair comparison (HANRAG-Fair, same base model as Adaptive-RAG) still shows gains on most benchmarks, which is the evidence the paper should lead with.

The soft spots are real, though. First, Table 1 reports SQuAD EM 39.80 with F1 39.76, and Natural Questions EM 56.40 with F1 49.12. EM can never exceed F1, instance-wise or on averages. These two cells are impossible, and the paper's claim that HANRAG 'achieves optimal results across all benchmark evaluation metrics' on single-hop rests on exactly that table. This is a load-bearing data error, not a typo to shrug off.

Second, the main comparison is not apples-to-apples: HANRAG uses Llama-3.1-8B while Adaptive-RAG uses FLAN-T5-XL. The HANRAG-Fair row narrows but does not close the gap, so the method probably does help, but the abstract's 'superior performance' overstates what the printed comparisons support.

Third, the compound benchmark is built by the authors, used for both training and testing, and not released. They claim no train/test overlap, but an external reader can't check the pipeline. The compound results should be treated as suggestive, not definitive.

Minor: no error bars or seeds, and a few references are placeholders (Natural Questions, TriviaQA). That's sloppy but fixable.

Bottom line: anyone building practical RAG systems for multi-hop QA should read this for the efficiency idea and the compound/complex distinction. It deserves a serious referee, but the referee should send it back for corrected Table 1, a headline fair comparison, released benchmark/code, and variance reporting. With those fixes, it could be a solid contribution.

Referee Report

4 major / 7 minor

Summary. The paper proposes HANRAG, a retrieval-augmented generation framework built around a fine-tuned 'Revelator' model that routes queries into four categories (straightforward, single-step, compound, and complex), decomposes compound queries into independent sub-queries, refines seed questions for complex queries, and filters retrieved documents by relevance before generation. The authors contribute a synthetic compound-question benchmark, use Qwen2-72B-instruct to generate training labels for routing, decomposition, refinement, relevance, and ending decisions, and evaluate on SQuAD, Natural Questions, TriviaQA, MuSiQue, HotpotQA, 2WikiMultihopQA, and their own compound benchmark. They report consistent gains over Adaptive-RAG, IRCoT, Self-RAG, and other baselines, with fewer retrieval steps, and include ablations and oracle-routing experiments.

Significance. If the reported results are correct, the paper would make a useful contribution by distinguishing compound from complex multi-hop queries, showing that asynchronous parallel retrieval for compound queries reduces retrieval steps, and demonstrating that relevance filtering improves robustness to noisy retrieved documents. The ablation study and oracle-routing analysis are informative, and the comparison spans several standard benchmarks. However, the central empirical claim is currently undermined by impossible metric pairs in Table 1, inconsistent step counts between Tables 2 and 6, and a main comparison that uses a stronger base LLM for HANRAG than for the key baseline; the self-built compound benchmark is also used for both training and evaluation. These issues must be resolved before the state-of-the-art claim is supported.

major comments (4)

[Table 1, Section 6] Table 1 reports HANRAG SQuAD EM=39.80 with F1=39.76 and Natural Questions EM=56.40 with F1=49.12. For token-level exact-match scoring, EM cannot exceed F1 for any individual prediction, and the inequality is preserved by averaging; the printed values therefore cannot be correct as computed. Because Section 6 uses Table 1 to assert that HANRAG 'achieves optimal results across all benchmark evaluation metrics' for single-hop datasets, this inconsistency is load-bearing. The authors should re-run the metric computation, report corrected values, and check the same evaluation pipeline for the multi-hop and compound tables.
[Tables 1 and 2; Section 5 (Implementation Details)] The headline comparison is asymmetric in base model: HANRAG uses Llama-3.1-8B-instruct for both the Revelator and the generator, while Adaptive-RAG uses FLAN-T5-XL (3B). The HANRAG-Fair rows in Tables 1 and 2, which use FLAN-T5-XL for HANRAG, show substantially smaller gains (e.g., SQuAD EM drops from 39.80 to 32.60 versus Adaptive-RAG's 26.80; TriviaQA EM is 52.40 versus Adaptive-RAG's 52.20). The main text should base its state-of-the-art claims on the fair variant or justify why the larger-model comparison is the appropriate one. As printed, the 'optimal across all metrics' claim conflates gains from HANRAG with gains from a stronger base LLM.
[Section 5, Appendix B.1, Section 4.2] The compound multi-hop benchmark is constructed by the authors using a Qwen2-72B-instruct pipeline, and the same benchmark's train split is used to train the router and decomposer while its test split is used for evaluation. A disjoint train/test split within a self-generated distribution does not rule out overfitting to the specific generation pipeline or topic distribution, so the reported 71.76 compound accuracy and 1.24 steps (Table 2) may not transfer to real compound queries. The paper should validate compound handling on an external or independently constructed compound-question set, or at minimum report a human evaluation and an analysis of distributional overlap.
[Table 6 vs Table 2; Section 6] The ablation table reports HANRAG's MuSiQue steps as 3.01, while Table 2 reports HANRAG's MuSiQue steps as 2.45 for what appears to be the same configuration. Since step reduction is a central efficiency claim, this discrepancy must be reconciled; if the ablation uses a different early-stopping or retriever setting, that should be stated explicitly. The same table also shows that removing the relevance discriminator lowers accuracy from 43.2 to 37.8 but changes EM by only 4.6 points, which deserves interpretation in the text.

minor comments (7)

[Section 4 heading] The section heading is spelled 'Methdology' and should read 'Methodology'.
[Algorithm 2, line 16] Algorithm 2 calls 'Revelator(q_i, D)' to produce the sub-answer, but elsewhere the retrieval-and-generation step is described as ANRAG; this is likely a typo but should be corrected for reproducibility.
[Appendix C] In Appendix C, 'teble 5' should be 'Table 5'.
[Tables 8 and 9] The case-study tables label Adaptive-RAG's handling of complex queries as 'Asynchronous retrieval', which conflicts with the paper's definition of complex queries as synchronous iterative retrieval; the terminology should be made consistent.
[Section 5 (Datasets)] Natural Questions and TriviaQA are cited as '(?)' in the datasets paragraph; proper references should be supplied.
[Section 4.2 vs Appendix B.2] Section 4.2 states 9,741 straightforward-query training samples for the router, while Appendix B.2 states 12,247; the discrepancy should be resolved.
[Algorithm 1] Algorithm 1's output line says 'Top 3 relevant passages for Q', but the algorithm returns LLM(D_rel), not the three passages; the output specification should be corrected.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: HANRAG's central claims are empirical comparisons on held-out external benchmarks, not outputs re-derived from fitted inputs.

full rationale

HANRAG's derivation chain is empirical rather than definitional. The Revelator is trained on routing, decomposition, refinement, relevance, and ending-discrimination data constructed from training splits (Section 4.2, Appendix B), and all headline results (Tables 1-2) are evaluated on held-out test splits of standard benchmarks (SQuAD, Natural Questions, TriviaQA, HotpotQA, 2WikiMultihopQA, MuSiQue) plus an author-built compound benchmark with explicitly disjoint train/dev/test samples. No parameter is fitted to the test-set metric that is then reported as a prediction; the ablation study (Appendix E) instead removes trained modules and measures the drop on MuSiQue, which is an independent check. Self-citations (e.g., HiRAG, PRGB) appear only as related-work pointers and are not load-bearing for the state-of-the-art claim. Two validity concerns do not amount to circularity: the author-built compound benchmark is used both to train the decomposer/router and to evaluate them, though on disjoint splits; and Table 1 reports HANRAG SQuAD EM 39.80 > F1 39.76 and NQ EM 56.40 > F1 49.12, which is impossible under the paper's own metric definitions and undermines the single-hop results as printed. These are correctness/reproducibility issues, not cases where a claimed prediction reduces by construction to its inputs, so the paper is not significantly circular.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper does not derive a mathematical result; it proposes an ML system. The free parameters are standard hyperparameters. The key assumptions are about the query taxonomy, the reliability of LLM-generated annotations, and the representativeness of the self-built benchmark. No new physical or conceptual entities are introduced beyond the named Revelator model, which is a trained software component rather than a scientific postulate.

free parameters (4)

top_k_retriever = 10
The retriever returns the top 10 documents for each query (Algorithm 1). This hyperparameter is chosen without reported tuning.
top_k_relevant = 3
The Revelator passes at most 3 relevant documents to the LLM (Algorithm 1 output). This number is fixed by hand.
max_steps = 5
The maximum number of retrieval-generation cycles is set to 5 for complex queries (Appendix E ablation).
learning_rate = 1e-4
LoRA fine-tuning uses initial learning rate 1e-4 with cosine schedule (Appendix D).

assumptions (4)

domain assumption The four-way query taxonomy (straightforward, single-step, compound, complex) is complete and sufficient for real-world multi-hop questions.
The framework routes every query into exactly one of these four classes, and the paper does not test robustness to out-of-taxonomy questions.
domain assumption Compound queries can be decomposed into independent sub-queries and answered with parallel retrieval.
Section 4.1 assumes asynchronous retrieval is valid for compound questions; misclassification would lead to missing dependencies.
ad hoc to paper Binary relevance labels from Qwen2-72B-instruct are reliable for training the relevance discriminator.
Section 4.2 uses an LLM to annotate <Q,D,IS_REL>, with no human validation or inter-annotator agreement.
ad hoc to paper The self-generated compound benchmark is representative of real compound queries.
The benchmark is built by prompting Qwen2-72B with Wikipedia documents and combining simple questions; its external validity is not established.

how reviews work

0 comments

Cite this review

Pith. "Pith review of HANRAG: Heuristic Accurate Noise-resistant Retrieval-Augmented Generation for Multi-hop Question Answering." pith.science (2026). https://pith.science/paper/CGVWIY2W

@misc{pith2026250909713,
  author       = {Pith},
  title        = {Pith review of: HANRAG: Heuristic Accurate Noise-resistant Retrieval-Augmented Generation for Multi-hop Question Answering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CGVWIY2W}},
  note         = {Machine review of arXiv:2509.09713}
}

read the original abstract

The Retrieval-Augmented Generation (RAG) approach enhances question-answering systems and dialogue generation tasks by integrating information retrieval (IR) technologies with large language models (LLMs). This strategy, which retrieves information from external knowledge bases to bolster the response capabilities of generative models, has achieved certain successes. However, current RAG methods still face numerous challenges when dealing with multi-hop queries. For instance, some approaches overly rely on iterative retrieval, wasting too many retrieval steps on compound queries. Additionally, using the original complex query for retrieval may fail to capture content relevant to specific sub-queries, resulting in noisy retrieved content. If the noise is not managed, it can lead to the problem of noise accumulation. To address these issues, we introduce HANRAG, a novel heuristic-based framework designed to efficiently tackle problems of varying complexity. Driven by a powerful revelator, HANRAG routes queries, decomposes them into sub-queries, and filters noise from retrieved documents. This enhances the system's adaptability and noise resistance, making it highly capable of handling diverse queries. We compare the proposed framework against other leading industry methods across various benchmarks. The results demonstrate that our framework obtains superior performance in both single-hop and multi-hop question-answering tasks.

Figures

Figures reproduced from arXiv: 2509.09713 by the authors.

**Figure 1.** Comparison of retrieval methods for Compound queries and Complex queries. A Complex query is composed of multiple sub-queries with strong logical reasoning relationships; in contrast, the sub-queries of a Compound query are almost independent. Synchronous retrieval is necessary for the former, while asynchronous retrieval is more efficient for the latter. effectively addressed. However, LLMs, which are trained on l… view at source ↗

**Figure 2.** Overall framework of HANRAG. The top-left section illustrates the functioning of the router, routing the query to the correct category. The bottom-left section outlines the workflow of ANRAG, which determines the relevance between the document and the query, ensuring that only noise-free documents are passed to the generation model. The right-hand section provides a detailed depiction of how HANRAG handles four diff… view at source ↗

**Figure 3.** Comparison of retrieval methods for Single and Complex queries. in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 5 canonical work pages

[1]

Straightforward question, which means that this question does not require external knowledge to be queried, and the information you know is enough to answer the question

work page
[2]

to answer it, but you only need to use external knowledge once

Single-step question, which means that the information you know cannot answer this question, and you need to use some external knowledge, such as searching the Internet, asking experts, etc. to answer it, but you only need to use external knowledge once

work page
[3]

Tiziano Labruna, Jon Ander Campos, and Gorka Azkune

Hirag: Hierarchical-thought instruction-tuning retrieval-augmented generation.arXiv preprint arXiv:2507.05714. Tiziano Labruna, Jon Ander Campos, and Gorka Azkune. 2024. When to retrieve: Teaching llms to utilize information retrieval effectively.Preprint, arXiv:2404.19705. Yanming Liu, Xinyue Peng, Xuhong Zhang, Weihao Liu, Jianwei Yin, Jiannan Cao, and ...

arXiv 2024
[4]

Jinming Nian, Zhiyuan Peng, Qifan Wang, and Yi Fang

Multi-hop question answering.Preprint, arXiv:2204.09140. Jinming Nian, Zhiyuan Peng, Qifan Wang, and Yi Fang

arXiv
[5]

OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, and Ilge Akkaya et al

W-rag: Weakly supervised dense retrieval in rag for open-domain question answering.Preprint, arXiv:2408.08444. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, and Ilge Akkaya et al. 2024. Gpt-4 technical report.Preprint, arXiv:2303.08774. Bhrij Patel, Vishnu Sashank Dorbala, Amrit Singh Bedi, and Dinesh Manocha. 2024. Multi-llm qa with em...

arXiv 2024
[6]

Preprint, arXiv:2401.15884

Corrective retrieval augmented generation. Preprint, arXiv:2401.15884. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Hao- ran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jianxin Yang, Jin Xu, Jingren Zhou, Jinze Bai, J...

arXiv 2024
[7]

needle-in-a- haystack

Llamafactory: Unified efficient fine-tuning of 100+ language models. InProceedings of the 62nd Annual Meeting of the Association for Compu- tational Linguistics (Volume 3: System Demonstra- tions), Bangkok, Thailand. Association for Computa- tional Linguistics. Chenghao Zhu, Nuo Chen, Yufei Gao, Yunyi Zhang, Prayag Tiwari, and Benyou Wang. 2024. Is your l...

arXiv 2024
[10]

Compound question, which means that this question is composed of multiple sub-questions, but these sub-questions are not related, or the correlation is relatively small, and no complex logical reasoning is required, but the information you know cannot answer the question, and it needs to be broken down into several sub-questions and then answered with the...

work page

Show all 12 references

[11]

Question

Complex question, which means that this question is composed of multiple sub-questions through complex logical nesting. There is a very strong logical relationship between these sub-questions. After decomposition, you still need to get the answer to a sub-question before you c...

1983
[1864]

Question

It has branches in Jacksonville, Florida, and Scottsdale, Arizona, as well as smaller clinics and hospitals in Minnesota, Iowa, and Wisconsin. It is consistently ranked as the best hospital in the world by major authoritative reports.“‘ Doc3: “‘On April 7, In 2015, Liu announc...

2015
[2024]

Chunjing Gan, Dan Yang, Binbin Hu, Hanxiao Zhang, Siyuan Li, Ziqi Liu, Yue Shen, Lin Ju, Zhiqiang Zhang, Jinjie Gu, Lei Liang, and Jun Zhou

Multi-llm text summarization.Preprint, arXiv:2412.15487. Chunjing Gan, Dan Yang, Binbin Hu, Hanxiao Zhang, Siyuan Li, Ziqi Liu, Yue Shen, Lin Ju, Zhiqiang Zhang, Jinjie Gu, Lei Liang, and Jun Zhou. 2024. Similarity is not all you need: Endowing retrieval augmented generation w...

2024 arXiv
[2025]

Lochan Basyal and Mihir Sanghvi

Reverse question answering: Can an llm write a question so hard (or bad) that it can’t answer? Preprint, arXiv:2410.15512. Lochan Basyal and Mihir Sanghvi. 2023. Text summa- rization using large language models: A comparative study of mpt-7b-instruct, falcon-7b-instruct, and o...

2023 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Straightforward question, which means that this question does not require external knowledge to be queried, and the information you know is enough to answer the question

work page

[2] [2]

to answer it, but you only need to use external knowledge once

Single-step question, which means that the information you know cannot answer this question, and you need to use some external knowledge, such as searching the Internet, asking experts, etc. to answer it, but you only need to use external knowledge once

work page

[3] [3]

Tiziano Labruna, Jon Ander Campos, and Gorka Azkune

Hirag: Hierarchical-thought instruction-tuning retrieval-augmented generation.arXiv preprint arXiv:2507.05714. Tiziano Labruna, Jon Ander Campos, and Gorka Azkune. 2024. When to retrieve: Teaching llms to utilize information retrieval effectively.Preprint, arXiv:2404.19705. Yanming Liu, Xinyue Peng, Xuhong Zhang, Weihao Liu, Jianwei Yin, Jiannan Cao, and ...

arXiv 2024

[4] [4]

Jinming Nian, Zhiyuan Peng, Qifan Wang, and Yi Fang

Multi-hop question answering.Preprint, arXiv:2204.09140. Jinming Nian, Zhiyuan Peng, Qifan Wang, and Yi Fang

arXiv

[5] [5]

OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, and Ilge Akkaya et al

W-rag: Weakly supervised dense retrieval in rag for open-domain question answering.Preprint, arXiv:2408.08444. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, and Ilge Akkaya et al. 2024. Gpt-4 technical report.Preprint, arXiv:2303.08774. Bhrij Patel, Vishnu Sashank Dorbala, Amrit Singh Bedi, and Dinesh Manocha. 2024. Multi-llm qa with em...

arXiv 2024

[6] [6]

Preprint, arXiv:2401.15884

Corrective retrieval augmented generation. Preprint, arXiv:2401.15884. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Hao- ran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jianxin Yang, Jin Xu, Jingren Zhou, Jinze Bai, J...

arXiv 2024

[7] [7]

needle-in-a- haystack

Llamafactory: Unified efficient fine-tuning of 100+ language models. InProceedings of the 62nd Annual Meeting of the Association for Compu- tational Linguistics (Volume 3: System Demonstra- tions), Bangkok, Thailand. Association for Computa- tional Linguistics. Chenghao Zhu, Nuo Chen, Yufei Gao, Yunyi Zhang, Prayag Tiwari, and Benyou Wang. 2024. Is your l...

arXiv 2024

[8] [10]

Compound question, which means that this question is composed of multiple sub-questions, but these sub-questions are not related, or the correlation is relatively small, and no complex logical reasoning is required, but the information you know cannot answer the question, and it needs to be broken down into several sub-questions and then answered with the...

work page