pith. sign in

arxiv: 2606.07074 · v1 · pith:EKUY7PZSnew · submitted 2026-06-05 · 💻 cs.LG · cs.AI

SlimSearcher: Training Efficiency-Aware Web Agents via Adaptive Reward Gating

Zequn Xie , Junjie Wang , Dan Yang , Jie Feng , Yue Shen , Jian Wang , Jinjie Gu This is my paper

Pith reviewed 2026-06-27 22:52 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords web agentsreinforcement learningtool use efficiencyreward shapingsupervised fine-tuningPareto efficiencyinformation retrieval agents
0
0 comments X

The pith

SlimSearcher reduces web agent tool calls by 17-58 percent while maintaining or improving accuracy on long tasks.

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

The paper sets out to train web research agents that solve complex information tasks without the wasteful tool calls and long trajectories common in accuracy-focused models. It does so by filtering training data during supervised fine-tuning to retain only successful yet economical trajectories. In the reinforcement learning phase it adds a reward mechanism that measures relative efficiency inside each sampled group of attempts and then applies a strict correctness check. The result is agents that complete benchmarks such as GAIA, BrowseComp, and XBenchDeepSearch with substantially fewer tool rounds. A sympathetic reader would care because lower tool use directly cuts the high computational expense of running these agents at scale.

Core claim

SlimSearcher pushes the Pareto frontier between accuracy and computational cost by applying Pareto-efficient filtration in the SFT stage to distill trajectories that are both successful and economical, and by introducing Adaptive Reward Gating in the RL stage, a mechanism that evaluates relative tool and token efficiency within a sampled cohort before cascading those metrics with a strict correctness gate to avoid brevity bias and reward hacking. Experiments on long-horizon benchmarks demonstrate reductions in average tool-call rounds of 17-58 percent while accuracy is maintained or improved.

What carries the argument

Adaptive Reward Gating, a dynamic reward-shaping mechanism that evaluates relative efficiency within cohorts and then applies a strict correctness gate.

If this is right

  • Average tool-call rounds drop 17-58 percent on GAIA, BrowseComp, and XBenchDeepSearch.
  • Accuracy stays the same or rises on those same benchmarks.
  • The efficiency gains appear in both the SFT and RL stages of training.
  • The gating approach limits reward hacking that absolute efficiency penalties often produce.

Where Pith is reading between the lines

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

  • The cohort-relative comparison may transfer to other reinforcement learning settings that involve variable-length trajectories.
  • Applying the same filtration-plus-gating pattern could reduce costs in agent domains outside web search, such as code or planning agents.
  • If the method scales, it would make repeated long-horizon agent runs more affordable for smaller labs or repeated experimentation.

Load-bearing premise

That measuring efficiency relative to other attempts in the same cohort and then requiring full correctness will stop the model from learning overly brief or hacked behaviors.

What would settle it

Training a new set of agents with SlimSearcher and observing no reduction in tool-call rounds or a drop in accuracy on a held-out long-horizon benchmark relative to standard training.

Figures

Figures reproduced from arXiv: 2606.07074 by Dan Yang, Jian Wang, Jie Feng, Jinjie Gu, Junjie Wang, Yue Shen, Zequn Xie.

Figure 1
Figure 1. Figure 1: Behavioral Analysis of the Efficiency Trap. (a) Blind Tool Dependency: The baseline agent indis￾criminately invokes external search tools for a common￾sense query resolvable via internal knowledge, leading to increased latency. (b) Performative Reasoning: As demonstrated in the complex query case (see [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: SlimSearcher: Multi-dimensional SFT filtering and adaptive reward gating enable web agents to learn [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of tool-call distribution and cumulative accuracy. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Cases of our synthesized data. temperature 1.0, while validation decoding adopts temperature 0.7, top-p of 0.8, and top-k of 20. Both actor and reference model FSDP configurations enable parameter offloading, and gradient check￾pointing is activated to reduce memory overhead. Ulysses sequence parallelism of size 8 is applied to the actor for efficient long-context training. The entire RL training pipeline … view at source ↗
Figure 5
Figure 5. Figure 5: The complete system prompt used to initialize the SlimSearcher agent. [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Reasoning trace comparison on GAIA. SlimSearcher solves in 4 phases with 22 tools; MiroThinker uses [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of reasoning traces on XBench. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of reasoning traces on XBench. SlimSearcher uses a strategic aggregation approach, while [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
read the original abstract

Deep research agents have demonstrated remarkable capabilities in complex information-seeking tasks, yet this power comes at a steep computational cost. Driven by accuracy-focused training paradigms, current models adopt brute-force strategies characterized by blind tool dependency and performative reasoning-generating long, redundant trajectories that are far from necessary for resolving these tasks, leading to wasteful tool calls and excessive token consumption. To overcome this efficiency trap, we propose SlimSearcher, a principled framework that pushes the Pareto frontier between accuracy and computational cost across both Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL). In the SFT stage, SlimSearcher employs Pareto-efficient filtration to distill trajectories that are both successful and economical, guiding the model toward inherently efficiency-aware search behaviors. During RL, we introduce Adaptive Reward Gating, a dynamic reward-shaping mechanism that evaluates relative tool and token efficiency within a sampled cohort. By cascading these adaptive efficiency metrics with a strict correctness gate, our approach effectively avoids the brevity bias associated with absolute penalties and mitigates reward hacking. Extensive experiments on long-horizon benchmarks, including GAIA, BrowseComp, and XBenchDeepSearch, demonstrate that SlimSearcher reduces average tool-call rounds by 17%-58% while maintaining or improving accuracy.

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 introduces SlimSearcher, a framework for training efficiency-aware web agents via two stages: (1) SFT with Pareto-efficient filtration to distill trajectories that are both successful and economical, and (2) RL with Adaptive Reward Gating, which evaluates relative tool/token efficiency within a sampled cohort and cascades these metrics with a strict correctness gate to avoid brevity bias and reward hacking. Experiments on long-horizon benchmarks (GAIA, BrowseComp, XBenchDeepSearch) claim 17%-58% reductions in average tool-call rounds while maintaining or improving accuracy.

Significance. If the results hold with proper validation, the work would meaningfully advance agent training by addressing the efficiency trap in accuracy-focused paradigms for deep research agents. The relative-metric approach in reward shaping offers a principled way to mitigate common RL issues like reward hacking, with potential for broader impact on practical deployment of web agents.

major comments (2)
  1. [Abstract] Abstract: The central empirical claim of 17%-58% tool-call reduction (with maintained accuracy) is reported without any baselines, error bars, dataset details, ablation results, or statistical tests, so the data-to-claim link cannot be evaluated.
  2. [RL stage] RL stage (Adaptive Reward Gating description): The claim that cascading cohort-relative efficiency metrics with a strict correctness gate avoids brevity bias and reward hacking is presented without formal analysis, proof of the property, or supporting ablations; this mechanism is load-bearing for the method's validity.
minor comments (1)
  1. The abstract would benefit from explicit definitions of the efficiency metrics (tool rounds, token consumption) and the exact Pareto filtration criteria used in SFT.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our work. We address each major comment point-by-point below, providing clarifications and committing to revisions that strengthen the manuscript without altering its core contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central empirical claim of 17%-58% tool-call reduction (with maintained accuracy) is reported without any baselines, error bars, dataset details, ablation results, or statistical tests, so the data-to-claim link cannot be evaluated.

    Authors: We agree that the abstract, as a high-level summary, does not embed the full experimental details. The manuscript body (Sections 4.1–4.3, Tables 1–3, and Appendix) contains the requested baselines (including comparisons to standard SFT/RL agents), error bars from multiple runs, dataset specifications for GAIA/BrowseComp/XBenchDeepSearch, ablation results, and statistical significance tests. To improve the abstract-to-evidence linkage, we will revise the abstract to explicitly reference the evaluation benchmarks and direct readers to the detailed experimental results and ablations. revision: partial

  2. Referee: [RL stage] RL stage (Adaptive Reward Gating description): The claim that cascading cohort-relative efficiency metrics with a strict correctness gate avoids brevity bias and reward hacking is presented without formal analysis, proof of the property, or supporting ablations; this mechanism is load-bearing for the method's validity.

    Authors: We acknowledge that the current description relies on design rationale and overall empirical gains rather than isolated formal analysis or component ablations. We will add a dedicated subsection in the RL stage (with new ablation tables) that isolates the correctness gate versus relative efficiency metrics, quantifies reductions in brevity bias and reward-hacking incidents across cohorts, and provides a step-by-step explanation of the cascading logic with supporting experimental evidence from our training runs. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The manuscript proposes SlimSearcher as a training framework using Pareto-efficient filtration during SFT and Adaptive Reward Gating (relative cohort metrics cascaded with a correctness gate) during RL. No equations, derivations, or first-principles results are shown that reduce by construction to fitted inputs or self-citations. Claims rest on empirical benchmark results (GAIA, BrowseComp, XBenchDeepSearch) rather than any self-referential loop. The central efficiency gains are presented as outcomes of the described procedure, not as tautological renamings or fitted predictions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no free parameters, axioms, or invented entities are stated in the provided text.

pith-pipeline@v0.9.1-grok · 5759 in / 940 out tokens · 34416 ms · 2026-06-27T22:52:54.594881+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

74 extracted references · 8 canonical work pages

  1. [1]

    arXiv preprint arXiv:2510.04618 , year=

    Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models , author=. arXiv preprint arXiv:2510.04618 , year=

    Pith/arXiv arXiv

  2. [2]

    arXiv preprint , year=

    DLER: Doing Length pEnalty Right - Incentivizing More Intelligence per Token via Reinforcement Learning , author=. arXiv preprint , year=

  3. [3]

    arXiv preprint arXiv:2602.14234 , year=

    REDSearcher: A Scalable and Cost-Efficient Framework for Long-Horizon Search Agents , author=. arXiv preprint arXiv:2602.14234 , year=

    arXiv

  4. [4]

    2026 , eprint=

    WebClipper: Efficient Evolution of Web Agents with Graph-based Trajectory Pruning , author=. 2026 , eprint=

    2026

  5. [5]

    Trends in cognitive sciences , volume=

    Cognitive offloading , author=. Trends in cognitive sciences , volume=. 2016 , publisher=

    2016

  6. [6]

    arXiv preprint arXiv:2503.23383 , year=

    Torl: Scaling tool-integrated rl , author=. arXiv preprint arXiv:2503.23383 , year=

    arXiv

  7. [7]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Webwalker: Benchmarking llms in web traversal , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  8. [8]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , year=

    WebVoyager: Building an End-to-End Web Agent with Large Multimodal Models , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , year=

  9. [9]

    arXiv preprint arXiv:2506.10055 , year=

    Taskcraft: Automated generation of agentic tasks , author=. arXiv preprint arXiv:2506.10055 , year=

    arXiv

  10. [10]

    arXiv preprint arXiv:2508.07976 , year=

    Beyond ten turns: Unlocking long-horizon agentic search with large-scale asynchronous rl , author=. arXiv preprint arXiv:2508.07976 , year=

    arXiv

  11. [11]

    arXiv preprint arXiv:2507.16812 , year=

    Megascience: Pushing the frontiers of post-training datasets for science reasoning , author=. arXiv preprint arXiv:2507.16812 , year=

    arXiv

  12. [12]

    arXiv preprint arXiv:2505.22648 , year=

    Webdancer: Towards autonomous information seeking agency , author=. arXiv preprint arXiv:2505.22648 , year=

    arXiv

  13. [13]

    Guo, Daya and Yang, Dejian and Zhang, Haowei and Song, Junxiao and Wang, Peiyi and Zhu, Qihao and Xu, Runxin and Zhang, Ruoyu and Ma, Shirong and Bi, Xiao and Zhang, Xiaokang and Yu, Xingkai and Wu, Yu and Wu, Z. F. and Gou, Zhibin and Shao, Zhihong and Li, Zhuoshu and Gao, Ziyi and Liu, Aixin and Xue, Bing and Wang, Bingxuan and Wu, Bochao and Feng, Bei ...

    work page doi:10.1038/s41586-025-09422-z

  14. [14]

    2025 , eprint=

    Kimi K2: Open Agentic Intelligence , author=. 2025 , eprint=

    2025

  15. [15]

    2025 , eprint=

    GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models , author=. 2025 , eprint=

    2025

  16. [16]

    Deep Research System Card , author =

  17. [17]

    Gemini Deep Research , author=

  18. [18]

    Claude Research , author=

  19. [19]

    Perplexity Reseaarch , author=

  20. [20]

    2025 , eprint=

    MiroThinker: Pushing the Performance Boundaries of Open-Source Research Agents via Model, Context, and Interactive Scaling , author=. 2025 , eprint=

    2025

  21. [21]

    2025 , eprint=

    Tongyi DeepResearch Technical Report , author=. 2025 , eprint=

    2025

  22. [22]

    2025 , eprint=

    WebLeaper: Empowering Efficiency and Efficacy in WebAgent via Enabling Info-Rich Seeking , author=. 2025 , eprint=

    2025

  23. [23]

    2025 , eprint=

    Step-DeepResearch Technical Report , author=. 2025 , eprint=

    2025

  24. [24]

    2025 , eprint=

    Can Pruning Improve Reasoning? Revisiting Long-CoT Compression with Capability in Mind for Better Reasoning , author=. 2025 , eprint=

    2025

  25. [25]

    Advances in Neural Information Processing Systems , year=

    A-mem: Agentic memory for llm agents , author=. Advances in Neural Information Processing Systems , year=

  26. [26]

    2025 , eprint=

    WebWeaver: Structuring Web-Scale Evidence with Dynamic Outlines for Open-Ended Deep Research , author=. 2025 , eprint=

    2025

  27. [27]

    2025 , eprint=

    TaskCraft: Automated Generation of Agentic Tasks , author=. 2025 , eprint=

    2025

  28. [28]

    2025 , eprint=

    WebDancer: Towards Autonomous Information Seeking Agency , author=. 2025 , eprint=

    2025

  29. [29]

    2025 , url =

    Open Deep Research , title =. 2025 , url =

    2025

  30. [30]

    2025 , url =

    GPT Research , title =. 2025 , url =

    2025

  31. [31]

    2025 , eprint=

    WebExplorer: Explore and Evolve for Training Long-Horizon Web Agents , author=. 2025 , eprint=

    2025

  32. [32]

    2025 , eprint=

    WebThinker: Empowering Large Reasoning Models with Deep Research Capability , author=. 2025 , eprint=

    2025

  33. [33]

    2025 , eprint=

    ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning , author=. 2025 , eprint=

    2025

  34. [34]

    and Zhang, Wen and Chen, Huajun

    Wang, Junjie and Chen, Mingyang and Hu, Binbin and Yang, Dan and Liu, Ziqi and Shen, Yue and Wei, Peng and Zhang, Zhiqiang and Gu, Jinjie and Zhou, Jun and Pan, Jeff Z. and Zhang, Wen and Chen, Huajun. Learning to Plan for Retrieval-Augmented Large Language Models from Knowledge Graphs. Findings of the Association for Computational Linguistics: EMNLP 2024...

    work page doi:10.18653/v1/2024.findings-emnlp.459 2024

  35. [35]

    Learning to Reason with

  36. [36]

    Guo, Daya and Yang, Dejian and Zhang, Haowei and Song, Junxiao and Zhang, Ruoyu and Xu, Runxin and Zhu, Qihao and Ma, Shirong and Wang, Peiyi and Bi, Xiao and others , journal=

  37. [37]

    Token-Budget-Aware LLM Reasoning

    Han, Tingxu and Wang, Zhenting and Fang, Chunrong and Zhao, Shiyu and Ma, Shiqing and Chen, Zhenyu. Token-Budget-Aware LLM Reasoning. Findings of the Association for Computational Linguistics: ACL 2025. 2025. doi:10.18653/v1/2025.findings-acl.1274

    work page doi:10.18653/v1/2025.findings-acl.1274 2025

  38. [38]

    2025 , eprint=

    Chain of Draft: Thinking Faster by Writing Less , author=. 2025 , eprint=

    2025

  39. [39]

    Brevity is the soul of sustainability: Characterizing LLM response lengths

    Poddar, Soham and Koley, Paramita and Misra, Janardan and Ganguly, Niloy and Ghosh, Saptarshi. Brevity is the soul of sustainability: Characterizing LLM response lengths. Findings of the Association for Computational Linguistics: ACL 2025. 2025. doi:10.18653/v1/2025.findings-acl.1125

    work page doi:10.18653/v1/2025.findings-acl.1125 2025

  40. [40]

    C o T -Valve: Length-Compressible Chain-of-Thought Tuning

    Ma, Xinyin and Wan, Guangnian and Yu, Runpeng and Fang, Gongfan and Wang, Xinchao. C o T -Valve: Length-Compressible Chain-of-Thought Tuning. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025. doi:10.18653/v1/2025.acl-long.300

    work page doi:10.18653/v1/2025.acl-long.300 2025

  41. [41]

    Self-Training Elicits Concise Reasoning in Large Language Models

    Munkhbat, Tergel and Ho, Namgyu and Kim, Seo Hyun and Yang, Yongjin and Kim, Yujin and Yun, Se-Young. Self-Training Elicits Concise Reasoning in Large Language Models. Findings of the Association for Computational Linguistics: ACL 2025. 2025. doi:10.18653/v1/2025.findings-acl.1289

    work page doi:10.18653/v1/2025.findings-acl.1289 2025

  42. [42]

    Stepwise Perplexity-Guided Refinement for Efficient Chain-of-Thought Reasoning in Large Language Models

    Cui, Yingqian and He, Pengfei and Zeng, Jingying and Liu, Hui and Tang, Xianfeng and Dai, Zhenwei and Han, Yan and Luo, Chen and Huang, Jing and Li, Zhen and Wang, Suhang and Xing, Yue and Tang, Jiliang and He, Qi. Stepwise Perplexity-Guided Refinement for Efficient Chain-of-Thought Reasoning in Large Language Models. Findings of the Association for Compu...

    work page doi:10.18653/v1/2025.findings-acl.956 2025

  43. [43]

    2025 , eprint=

    O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning , author=. 2025 , eprint=

    2025

  44. [44]

    2025 , eprint=

    L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning , author=. 2025 , eprint=

    2025

  45. [45]

    C oncise RL : Conciseness-Guided Reinforcement Learning for Efficient Reasoning Models

    Dumitru, Razvan-Gabriel and Peteleaza, Darius and Yadav, Vikas and Pan, Liangming. C oncise RL : Conciseness-Guided Reinforcement Learning for Efficient Reasoning Models. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025. doi:10.18653/v1/2025.findings-emnlp.927

    work page doi:10.18653/v1/2025.findings-emnlp.927 2025

  46. [46]

    2023 , html =

    Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , booktitle =. 2023 , html =

    2023

  47. [47]

    arXiv preprint arXiv:2504.12516 , year=

    Browsecomp: A simple yet challenging benchmark for browsing agents , author=. arXiv preprint arXiv:2504.12516 , year=

    Pith/arXiv arXiv

  48. [48]

    arXiv preprint arXiv:2501.14249 , year=

    Humanity's last exam , author=. arXiv preprint arXiv:2501.14249 , year=

    Pith/arXiv arXiv

  49. [49]

    The Twelfth International Conference on Learning Representations , year=

    Gaia: a benchmark for general ai assistants , author=. The Twelfth International Conference on Learning Representations , year=

  50. [50]

    Xbench-DeepSearch , author =

  51. [51]

    SerpAPI: Google Search API , author =

  52. [52]

    2025 , eprint=

    Qwen3 Technical Report , author=. 2025 , eprint=

    2025

  53. [53]

    2025 , eprint=

    DeepSeek-V3 Technical Report , author=. 2025 , eprint=

    2025

  54. [54]

    Introducing Claude 4 , author =

  55. [55]

    Introducing OpenAI o3 and o4-mini , author =

  56. [56]

    Machine learning , volume=

    F*: an interpretable transformation of the F-measure , author=. Machine learning , volume=. 2021 , publisher=

    2021

  57. [57]

    arXiv preprint arXiv:2510.18939 , year=

    Lost in the Maze: Overcoming Context Limitations in Long-Horizon Agentic Search , author=. arXiv preprint arXiv:2510.18939 , year=

    arXiv

  58. [58]

    arXiv preprint arXiv:2312.10003 , year=

    Rest meets react: Self-improvement for multi-step reasoning llm agent , author=. arXiv preprint arXiv:2312.10003 , year=

    arXiv

  59. [59]

    arXiv preprint arXiv:2503.09516 , year=

    Search-r1: Training llms to reason and leverage search engines with reinforcement learning , author=. arXiv preprint arXiv:2503.09516 , year=

    Pith/arXiv arXiv

  60. [60]

    Self-dc: When to reason and when to act? self divide-and-conquer for compositional unknown questions , author=. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    2025

  61. [61]

    arXiv preprint arXiv:2507.02592 , year=

    Websailor: Navigating super-human reasoning for web agent , author=. arXiv preprint arXiv:2507.02592 , year=

    Pith/arXiv arXiv

  62. [62]

    Webshaper: Agentically data synthesizing via information-seeking formalization, 2025 , author=

    2025

  63. [63]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Cot-valve: Length-compressible chain-of-thought tuning , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  64. [64]

    URL https://arxiv

    Chain of draft: Thinking faster by writing less, 2025 , author=. URL https://arxiv. org/abs/2502.18600 , year=

    arXiv 2025

  65. [65]

    Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    Token-budget-aware llm reasoning , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  66. [66]

    Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    Self-training elicits concise reasoning in large language models , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  67. [67]

    URL https://arxiv

    L1: Controlling how long a reasoning model thinks with reinforcement learning, 2025 , author=. URL https://arxiv. org/abs/2503.04697 , volume=

    Pith/arXiv arXiv 2025

  68. [68]

    arXiv preprint arXiv:2504.21776 , year=

    Webthinker: Empowering large reasoning models with deep research capability , author=. arXiv preprint arXiv:2504.21776 , year=

    Pith/arXiv arXiv

  69. [69]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    Pith/arXiv arXiv

  70. [70]

    arXiv preprint arXiv:2402.03300 , year=

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models , author=. arXiv preprint arXiv:2402.03300 , year=

    Pith/arXiv arXiv

  71. [71]

    Proceedings of the 29th symposium on operating systems principles , pages=

    Efficient memory management for large language model serving with pagedattention , author=. Proceedings of the 29th symposium on operating systems principles , pages=

  72. [72]

    Proceedings of the Twentieth European Conference on Computer Systems , pages=

    Hybridflow: A flexible and efficient rlhf framework , author=. Proceedings of the Twentieth European Conference on Computer Systems , pages=

  73. [73]

    2024 , eprint=

    SWIFT:A Scalable lightWeight Infrastructure for Fine-Tuning , author=. 2024 , eprint=

    2024

  74. [74]

    2025 , howpublished=

    rLLM: A Framework for Post-Training Language Agents , author=. 2025 , howpublished=

    2025