Event-Grounded Question Answering over Long Audio via Structured Retrieval

Arvind Krishna Sridhar; Erik Visser; Kartik Hegde; Naveen Vakada; Yinyi Guo

REVIEW 4 major objections 6 minor 7 references

Event-Grounded Question Answering over Long Audio via Structured Retrieval

T0 review · 4 major / 6 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read Answering questions over multi-hour audio can be done accurately and in under a second by first converting the audio into timestamped event logs and retrieving only the events a query needs.

desk verdict A sensible long-audio QA architecture whose headline accuracy numbers are circular — the abstract also claims results the body never reports. read the letter →

arxiv 2602.14612 v5 pith:HWO4QPRL submitted 2026-02-16 eess.AS cs.AIcs.LG

Kartik Hegde , Arvind Krishna Sridhar , Naveen Vakada , Yinyi Guo , Erik Visser This is my paper

classification eess.AScs.AIcs.LG

keywords long-audioQAaudiogroundingmodelretrieval-augmentedgenerationeventdetectiontemporaltext-to-SQLstructuredretrievaledge-clouddeployment

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 show that natural-language question answering over multi-hour audio does not require feeding the whole recording to a large audio-language model. Instead, the system converts audio into timestamped event records using an open-vocabulary grounding model, stores them in a SQL database, and answers each query by resolving time references, classifying intent, retrieving only the relevant events, and letting a compact LLM generate a grounded response. On synthetic 24-hour Home and Industrial IoT benchmarks, this structured pipeline reaches 76.88% and 71.10% overall accuracy with average query latencies below 0.6 seconds, substantially outperforming vanilla retrieval-augmented generation (RAG) and text-to-SQL baselines. The paper also shows that feeding this structured event metadata to existing audio-language models improves their temporal detection F1 by 11 to 17 percentage points. The significance is a practical, deployment-oriented route to long-audio QA that is cheaper, faster, and more temporally precise than purely generative approaches.

What carries the argument

The Audio Grounding Model (AGM) — an open-vocabulary text-to-audio grounding network that outputs timestamped event records (event name, start/end times, confidence, loudness) for free-form sound queries. These logs feed a SQL event database; at query time, a pipeline of rephrasing, time resolution, intent classification, and embedding-based top-k retrieval selects the evidence that the LLM must use to answer.

What would settle it

Take a 24-hour recording with an independent human-annotated set of all events (not derived from the AGM), ask the system questions about events the AGM missed, and check whether it answers them correctly; if it fails on those questions, the accuracy numbers do not reflect true audio understanding. A simpler version: compute LA-RAG's recall of human-annotated events and see whether it matches the detection accuracy reported.

Watch

Extended reading notes

Core claim

The central claim is that explicit timestamped grounding via structured event logs, followed by intent-aware retrieval and LLM generation, yields large accuracy gains over both retrieval-augmented generation on raw event text and text-to-SQL over the database. Detection, counting, and summary tasks all improve, and the system stays fast because audio grounding runs once at index time on edge hardware, while only filtered event records are sent to a cloud-hosted LLM. The paper attributes these gains to three design choices: the AGM produces well-structured event logs; a time-resolution module maps varied natural-language expressions (12-hour, 24-hour, shift references, before/after, durations

Load-bearing premise

The evaluation's ground truth for Simple-QA pairs comes from the AGM's own event annotations, so questions only exist about events the detector found; any event the AGM misses cannot be probed, and the reported accuracy is not an independent measure of audio understanding.

Editorial extensions

If this is right

Long-audio QA can be answered accurately without ingesting raw audio into the model, sidestepping context-length limits of current audio-language models.
Sub-second query latency is achievable even for 24-hour recordings when event extraction happens offline at index time.
Mid-sized LLMs (roughly 7–14B active parameters) are sufficient when given structured, timestamped evidence; very large models are not required.
Adding structured retrieval metadata improves the temporal detection F1 of existing audio-language models by 11–17%, showing event logs complement generative models.
A hybrid edge–cloud deployment keeps raw audio local for privacy while transmitting only compact event metadata to the cloud LLM.

Reading between the lines

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

Because the Simple-QA ground truth is generated from the AGM's own event detections, the reported accuracy measures consistency with the AGM's event log, not independent acoustic understanding; the AGM's detection quality is therefore the true ceiling of the whole system.
The evaluation uses a predefined closed set of sound classes, so open-vocabulary performance in unrestricted acoustic environments remains untested; real-world deployments with novel sounds could degrade.
The synthetic benchmark concatenates short labeled clips with template-generated questions, which may yield higher accuracy than naturally continuous audio with human-annotated events; a real-world corpus would be a stronger test.
The same 'convert raw signal to structured events first, then reason' pattern could extend to other long modalities such as video or sensor streams, where context-length limits similarly block direct modeling.

Share X Bluesky LinkedIn Reddit HN

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

Colleague — quick take on arXiv:2602.14612. The system is plausible; the evaluation is not. The headline accuracy numbers (76.88% / 71.10%) come from questions built from the AGM's own event annotations, so they measure how well the LLM reads the AGM's log, not how well the system understands audio. That circularity is the load-bearing flaw, and the reader's report has it right.

What the paper does well: it puts together a sensible stack — open-vocabulary SED (WSTAG-style), SQL event store, time-expression resolution, intent routing, LLM generation — and shows it running on edge hardware with low latency. The relative comparison of query strategies (LA-RAG vs RAG vs Text2SQL) is informative because all baselines share the same event log; it tells you structured retrieval helps downstream reasoning. The model-scaling table and time-resolution ablation are useful. The DESED F1 of 74.8 gives some independent evidence the detector is not garbage. The limitations section is honest about propagation of AGM errors.

Soft spots in proportion: the circular QA generation is severe. Complex-QA pairs exist but the headline table only reports Simple-QA. The abstract promises CASTELLA-QA and 11-17% F1 gains for LALMs, and these are nowhere in the text — that is an unacceptable abstract-body mismatch. No code or data, and the system is proprietary, so others cannot re-run it. The human eval is small (113 pairs) with moderate GPT-4o correlation. Still, the core architecture could be validated with independent ground truth (e.g., human-annotated events or a public SED dataset re-purposed for QA). It's fixable.

Who gets value: researchers and engineers working on practical long-audio QA, and anyone teaching evaluation design — this is a good case of ground truth contamination. It deserves a serious referee, but the referee should be instructed to check the abstract against the body and demand either independent QA labels or a reframing of the accuracy numbers as 'event-log QA accuracy.' My own verdict is skeptical, but not because the idea is bad — because the evidence as presented does not support the claims as stated.

Referee Report

4 major / 6 minor

Summary. The paper presents LA-RAG, a hybrid system for natural-language QA over multi-hour audio. An Audio Grounding Model (AGM) converts audio into timestamped event records; these are stored in a SQLite database. At query time, the system performs query rephrasing, temporal reference resolution, intent classification, embedding/top-k retrieval, and LLM-based answer generation. The authors construct synthetic 24-hour Home-IoT and Industrial-IoT benchmarks with template-based QA pairs, compare against RAG and text-to-SQL baselines, and report overall accuracies of 76.88% and 71.10% with sub-0.6s latency. They also claim that structured retrieval improves accuracy, that CASTELLA was augmented with QA pairs, and that LALMs achieve temporal F1 gains of 11-17% when augmented with their retrieval metadata.

Significance. If independently validated, the core idea is practically significant: event-level structured logs combined with LLM reasoning could offer a low-latency, cost-effective alternative to long-context audio-language models. The paper has concrete strengths: a modular edge-cloud deployment with latency measurements, a sweep across LLM model scales, an external SED evaluation of the AGM on the AudioMarathon/DESED benchmark (74.8 F1), a small human evaluation, and a reasonably detailed system description. However, the evaluation is currently circular, some headline results advertised in the abstract are absent from the body, and the benchmark is synthetic and closed-set. The significance is therefore conditional on substantially stronger evidence than the paper currently provides.

major comments (4)

[Section 3.1, Table 1] The QA evaluation is circular. Simple-QA pairs are 'created deterministically from ground-truth event annotations produced by the Audio Grounding Model module.' The same AGM event log is the content of the SQL database that LA-RAG retrieves from. Thus a missed AGM event cannot be probed by any question, and an AGM false positive becomes the 'correct' answer. The reported 76.88% / 68.92% accuracies in Table 1 measure the LLM's ability to read the AGM's own event log, not independent audio understanding. The AGM's external DESED F1 of 74.8 (Appendix C) is not connected to QA accuracy. Appendix E's acknowledged reliance on AGM detection quality does not mitigate this design. This undermines the paper's central empirical claim.
[Abstract vs. full text] The abstract advertises two results that do not appear anywhere in the body: (i) augmentation of CASTELLA with QA pairs, and (ii) LALM temporal-detection F1 gains of 11-17% when augmented with structured retrieval metadata. No section, table, or appendix reports a CASTELLA-QA experiment or an LALM F1 comparison. These are load-bearing advertised results, not peripheral. Either the experiments must be added with full evaluation details, or the claims must be removed from the abstract.
[Tables 1 and 2, Abstract] The headline accuracy is ambiguous. The abstract says LA-RAG achieves 71.10% on Industrial-IoT, but Table 1 reports 68.92% for AGM LA-RAG under what appears to be the default Phi-4-mini configuration (Sec. 4.9). The 71.10% value appears only in Table 2 as the Phi-3.5-MoE result. The paper should specify which model configuration underlies the abstract's headline and why Table 1 and Table 2 report different Industrial-IoT numbers for 'LA-RAG.' As written, the reader cannot tell which result is being advertised.
[Section 4.1] The synthetic benchmark further limits the evidence. The acoustic classes are restricted to predefined lists, the same closed set is passed to the downstream prompts, and the QA pairs are generated from the same annotation process that produces the database records. This creates a favorable closed-world setting: the LLM is told the answer vocabulary and the database was built from the same event definitions. This may substantially inflate accuracy relative to open-vocabulary, real-world long-audio QA. The authors acknowledge in Appendix E that the benchmark 'does not fully capture the diversity of real-world acoustic environments,' but the evaluation design does not address this concern.

minor comments (6)

[Section 4.7 / Table 3] The time-resolution evaluation uses only 45 QA pairs, and the hard-condition accuracy is 30% for the combined system. Report more instances, confidence intervals, and an error breakdown by temporal-expression type.
[Section 4.4] The AGM decision threshold (0.8) and median-filter window (0.3s) are held constant across all experiments. A sensitivity analysis would help assess how robust the downstream QA accuracy is to these choices.
[Section 4.3] The RAG baseline uses a fixed k=5, and the AF3-based variant is prompted to produce structured outputs that 'fail to produce reliably structured outputs.' No retrieval-recall numbers are reported, so it is unclear whether the RAG baselines are strong or artificially weak.
[Appendix D] The human evaluation uses only 11 summarization questions, making the category mean unstable. Report per-item variance or confidence intervals, and consider increasing the sample for the summary category.
[Sections 1, 4.8, F, G] Reproducibility is limited: the code, data, and benchmark are not released, and the system is described as proprietary. Contribution (ii) claims 'a complete and reproducible implementation stack,' but no code or data is provided. The first-person note in Section 1 ('We cannot provide a demo link...') is also unusually informal for a journal submission.
[General] Figure 1 is referenced without a descriptive caption; Figures 3 and 4 are screenshots that add little. Some citations are inconsistent (e.g., 'Arah A. 2024' for Phi-3). These are presentation issues only.

Circularity Check

1 steps flagged · score 6.0 of 10

Evaluation is circular: Simple-QA ground truth is generated from the AGM's own event annotations, and LA-RAG answers from the same AGM event log, so reported accuracy measures log-consistency rather than independent audio QA.

self definitional [Section 3.1 (Automated QA generation); Section 3.2/3.2.1; Tables 1-2]
"Simple-QA pairs are created deterministically from ground-truth event annotations produced by the Audio Grounding Model (AGM) module, covering detection, counting, and summary queries... The output of this module is stored in JSON format which is then inserted into a SQLite database for question answering."

The same AGM outputs that define the evaluation ground truth are the only event records inserted into the SQL database that LA-RAG queries and reasons over. Therefore an AGM false negative removes the corresponding question from the test set, and an AGM false positive becomes the officially correct answer. The Table 1/2 accuracies (76.88%/71.10%) thus measure the LLM's ability to read and aggregate the AGM's own log after time resolution and intent classification, not the system's audio understanding; the external DESED F1=74.8 (Appendix C) is a separate AGM-only result and does not validate the QA numbers.

full rationale

The single load-bearing circular step is the QA benchmark generation. Section 3.1 states that Simple-QA pairs are created deterministically from ground-truth event annotations produced by the Audio Grounding Model (AGM) module, and Section 3.2 inserts exactly those AGM outputs into the SQLite database queried by LA-RAG. Consequently, no question can exist for an event the AGM missed, and any false positive is enshrined as correct, so the absolute accuracy numbers are not an independent measure of audio understanding. The paper itself acknowledges in Appendix E that missed or noisy detections can propagate to downstream reasoning. However, this is a partial circularity, not a total one: Table 1's comparison among AGM+RAG, AGM+Text2SQL, and AGM+LA-RAG holds the event log fixed and tests retrieval/query-processing choices, which is independently meaningful and supports the narrower structured-retrieval claim. No load-bearing self-citation or imported-uniqueness chain is present: AGM training follows the external WSTAG framework (Xu et al., 2024), and DESED/AudioMarathon provide an external AGM check. Separately, the abstract's CASTELLA-QA augmentation and 11-17% LALM F1 gains do not appear anywhere in the body (no CASTELLA-QA table, no LALM results), which is an omitted-support/reproducibility concern, not a circular one; it does not increase the circularity score but should weigh on overall confidence. Given one central evaluation quantity reduces by construction to the system's own event log, a score of 6 is appropriate.

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

The central claim rests on the AGM's event log; its construction (thresholds, WSTAG training) and the definition of QA ground truth from that log are the main unaccounted inputs.

free parameters (4)

AGM decision threshold = 0.8
Global threshold applied to framewise grounding scores to produce event boundaries (Section 4.4); directly determines the event log that feeds all QA results.
AGM median filter window = 0.3 s
Post-processing median filter on framewise scores (Section 4.4); shapes event boundaries and counts.
RAG retrieval k = 5
Number of retrieved snippets used for all RAG baselines (Section 4.3); affects baseline comparisons.
Synthetic benchmark SNR / loudness = 6 dB SNR, -16 LUFS
Construction parameters for the 24-hour concatenated audio (Section 4.1); affects difficulty and realism.

assumptions (4)

domain assumption AGM trained with WSTAG on AudioCaps yields open-vocabulary event detections sufficient for QA
The entire pipeline relies on AGM event logs; only validated externally on DESED SED (F1=74.8), not on the QA benchmarks.
ad hoc to paper QA pairs generated from AGM event annotations provide ground truth for question answering
Section 3.1 constructs Simple-QA from AGM-produced annotations; this imports system output as ground truth and is the main circularity.
ad hoc to paper Synthetic concatenated audio with closed-set classes approximates real-world long-audio queries
Section 4.1 restricts acoustic classes to predefined lists and generates template QA; no real multi-hour recordings evaluated in the full text.
domain assumption GPT-4o judge with category rubrics is a valid proxy for human assessment
Section 4.2 / Appendix D; human-GPT per-item Pearson correlation is 0.57, only moderate.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Event-Grounded Question Answering over Long Audio via Structured Retrieval." pith.science (2026). https://pith.science/paper/HWO4QPRL

@misc{pith2026260214612,
  author       = {Pith},
  title        = {Pith review of: Event-Grounded Question Answering over Long Audio via Structured Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HWO4QPRL}},
  note         = {Machine review of arXiv:2602.14612}
}

read the original abstract

Answering natural-language questions over multi-hour audio requires both event recognition and temporal grounding. Current large audio-language models perform well on short clips, but are limited by context length, query-time cost, and weak temporal localization. We present LA-RAG (Long Audio-Retrieval Augmented Generation), a structured framework that converts continuous audio into timestamped event records using an open-vocabulary Audio Grounding Model (AGM), stores them in a SQL event database, and answers queries through intent-aware retrieval followed by LLM-based generation. LA-RAG supports offline grounding mode, where long recordings are pre-indexed for low-latency QA, and inference-time grounding mode, where query-conditioned grounding is performed for shorter open-ended clips. We create 24-hour Home-IoT and Industrial-IoT audio benchmarks and augment CASTELLA, a real-world audio moment retrieval dataset with QA pairs. In offline grounding mode, LA-RAG achieves 76.88% overall accuracy on Home-IoT and 71.10% on Industrial-IoT, with average query latencies below 0.6 seconds. In inference-time grounding mode, state-of-the-art LALMs achieve competitive event-detection accuracy on CASTELLA-QA but low temporal detection F1. We further show that LALMs augmented with our structured retrieval metadata achieve consistent temporal detection improvements, with F1 gains of 11-17% across baseline models with improved latency. These results show that explicit timestamped grounding and structured retrieval provide a practical complement to generative audio-language models for deployment-oriented long-audio QA.

Figures

Figures reproduced from arXiv: 2602.14612 by the authors.

**Figure 1.** Chat Example for LongAudio-RAG manual review impractical and motivating systems that can answer natural-language questions about events and their timing. Unlike short-clip audio tasks, long audio question answering (QA) must handle time-bounded queries, aggregate counts, and narrative summaries, requiring precise temporal grounding and hallucination-free language generation. Recent progress in audio-language modeli… view at source ↗

**Figure 2.** LongAudio-RAG (LA-RAG): Proposed method for long audio question answering. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Software Implementation Architecture [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: LongAudio-RAG user interface service runs a compact PyTorch model via FastAPI, producing JSON event logs with temporal metadata and supporting custom sound enrollment by allowing users to register new classes from example recordings. Its low computational footprint en…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

7 extracted references · 5 linked inside Pith

[2]

InProceedings of the 2021 Conference of the North American Chapter of the Association for Computa- tional Linguistics: Human Language Technologies, pages 1337–1350

Structure-grounded pretraining for text-to-sql. InProceedings of the 2021 Conference of the North American Chapter of the Association for Computa- tional Linguistics: Human Language Technologies, pages 1337–1350. Haytham M Fayek and Justin Johnson. 2020. Temporal reasoning via audio question answering.IEEE/ACM Transactions on Audio, Speech, and Language P...

2021
[5]

An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others

Towards weakly supervised text-to-audio grounding.IEEE Transactions on Multimedia. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others
[7]

Zijia Zhao, Longteng Guo, Tongtian Yue, Sihan Chen, Shuai Shao, Xinxin Zhu, Zehuan Yuan, and Jing Liu

Qwen2.5 technical report.Preprint, arXiv:2412.15115. Zijia Zhao, Longteng Guo, Tongtian Yue, Sihan Chen, Shuai Shao, Xinxin Zhu, Zehuan Yuan, and Jing Liu. 2023. Chatbridge: Bridging modalities with large language model as a language catalyst.arXiv preprint arXiv:2305.16103. Ethical Considerations:This work involves pro- cessing long-duration audio, which...

arXiv 2023
[2019]

On efficient retrieval of top similarity vectors. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5236–5246. Xichen Tan, Yunfan Ye, Yuanjing Luo, Qian Wan, Fang Liu, and Zhiping Cai. 2025. Rag-adapter: A plug- and-play r...

arXiv 2019
[2021]

Is space-time attention all you need for video understanding? InProceedings of the 38th Inter- national Conference on Machine Learning (ICML). PMLR. Joao Carreira and Andrew Zisserman. 2017. Quo vadis, action recognition? a new model and the kinetics dataset. Inproceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 6299–6308....

arXiv 2017
[2024]

InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6288–6313

Gama: A large audio-language model with ad- vanced audio understanding and complex reasoning abilities. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6288–6313. Arushi Goel, Sreyan Ghosh, Jaehyeon Kim, Sonal Ku- mar, Zhifeng Kong, Sang-gil Lee, Chao-Han Huck Yang, Ramani Duraiswami, Dinesh Manocha, Rafael ...

arXiv 2024
[2025]

An Yang, Baosong Yang, Beichen Zhang, and et al

Qwen3 technical report.arXiv preprint arXiv:2505.09388. An Yang, Baosong Yang, Beichen Zhang, and et al

arXiv

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [2]

InProceedings of the 2021 Conference of the North American Chapter of the Association for Computa- tional Linguistics: Human Language Technologies, pages 1337–1350

Structure-grounded pretraining for text-to-sql. InProceedings of the 2021 Conference of the North American Chapter of the Association for Computa- tional Linguistics: Human Language Technologies, pages 1337–1350. Haytham M Fayek and Justin Johnson. 2020. Temporal reasoning via audio question answering.IEEE/ACM Transactions on Audio, Speech, and Language P...

2021

[2] [5]

An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others

Towards weakly supervised text-to-audio grounding.IEEE Transactions on Multimedia. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others

[3] [7]

Zijia Zhao, Longteng Guo, Tongtian Yue, Sihan Chen, Shuai Shao, Xinxin Zhu, Zehuan Yuan, and Jing Liu

Qwen2.5 technical report.Preprint, arXiv:2412.15115. Zijia Zhao, Longteng Guo, Tongtian Yue, Sihan Chen, Shuai Shao, Xinxin Zhu, Zehuan Yuan, and Jing Liu. 2023. Chatbridge: Bridging modalities with large language model as a language catalyst.arXiv preprint arXiv:2305.16103. Ethical Considerations:This work involves pro- cessing long-duration audio, which...

arXiv 2023

[4] [2019]

On efficient retrieval of top similarity vectors. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5236–5246. Xichen Tan, Yunfan Ye, Yuanjing Luo, Qian Wan, Fang Liu, and Zhiping Cai. 2025. Rag-adapter: A plug- and-play r...

arXiv 2019

[5] [2021]

Is space-time attention all you need for video understanding? InProceedings of the 38th Inter- national Conference on Machine Learning (ICML). PMLR. Joao Carreira and Andrew Zisserman. 2017. Quo vadis, action recognition? a new model and the kinetics dataset. Inproceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 6299–6308....

arXiv 2017

[6] [2024]

InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6288–6313

Gama: A large audio-language model with ad- vanced audio understanding and complex reasoning abilities. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6288–6313. Arushi Goel, Sreyan Ghosh, Jaehyeon Kim, Sonal Ku- mar, Zhifeng Kong, Sang-gil Lee, Chao-Han Huck Yang, Ramani Duraiswami, Dinesh Manocha, Rafael ...

arXiv 2024

[7] [2025]

An Yang, Baosong Yang, Beichen Zhang, and et al

Qwen3 technical report.arXiv preprint arXiv:2505.09388. An Yang, Baosong Yang, Beichen Zhang, and et al

arXiv