CB-Orchestrator is an adaptive framework for optimizing large language model (LLM) agent workflows. It separates workflow generation—via evolutionary search—from workflow selection, which is treated as a contextual bandit problem to enable sample‑efficient learning. The approach reduces training‑phase API token usage by up to 40.55% and overall token overhead by at least 29.73%, while outperforming baselines on five benchmarks, including the MATH dataset.
By Jiahang Sun, Zhiwei Shang, Zhipiao Liu, Hongwei Yang, Guoqing Xie, Shuang Qiu, Zhongxiang Dai
arXiv:2607. 29241v1 Announce Type: cross Abstract: Optimizing modern recommender models still depends heavily on engineers manually iterating over architectural, objective, and training-strategy changes.
By Haoran Ling, Yuecheng Li, Zeyu Song, Jing Yao, Shuwen Kang, Chi Lu, Wenjin Wu, Peng Jiang
COBALT is a contextual bandit learning method that reuses offline multi-turn code generation trajectories to train large language models efficiently. It treats multi-turn code generation as a one-step recoverable Markov decision process, collecting partial trajectories from a reference LLM and training the target model to complete each partial trajectory in a single step. Experiments on LiveCodeBench show that COBALT improves Pass@1 scores for R1-Distill 8B and Qwen3 8B by up to 9.0 and 6.2 points, matches the performance of two online RL baselines, and enhances robustness against misleading execution feedback.
By Ziru Chen, Dongdong Chen, Ruinan Jin, Yingbin Liang, Yujia Xie, Huan Sun
In-Context Batch Black-box Optimization via Flow Matching introduces FMTR, an amortized method that learns a continuous, history-conditioned proposal distribution for variable-size batch sampling in expensive black-box optimization settings. FMTR is trained offline using conditional flow matching on Thompson-sampling argmaxes refined through a horizon-dependent Langevin pushforward map, enabling adaptive exploration-exploitation over the optimization horizon. At test time, FMTR generates batches without task-specific surrogate refitting or numerical acquisition optimization, achieving strong performance on synthetic and real-world benchmarks while significantly reducing batch selection computational cost.
By Quoc-Anh Hoang Nguyen, Xinyu Zhang, Daolang Huang, Julien Martinelli, Samuel Kaski
arXiv:2609.06670v1 Announce Type: new
Abstract: The In-context learning (ICL) paradigm aids large language models (LLMs) to adapt to new tasks without need for fine-tuning. However, selecting an opti...
By V Venktesh, Cem levi, Avishek Anand
arXiv:2508. 12116v2 Announce Type: replace-cross Abstract: As numerous instruction-tuning datasets continue to emerge, dynamically balancing and optimizing their mixtures has become a critical challenge.
By Haebin Shin, Lei Ji, Xiao Liu, Zhiwei Yu, Hyunwoo Yoo, Qi Chen, Yeyun Gong
arXiv:2606. 07726v1 Announce Type: new Abstract: Large Language Models are typically benchmarked by evaluating every model on every test query.
By Zifan Lyu, Chahine Nejma, Tobias Wegel, Fanny Yang, Florian E. Dorner
COBRA‑Skills is a new framework that treats skill optimization for large language model agents as a budgeted sequential problem over a dynamically evolving candidate set. It uses contextual‑bandit prioritization to focus evaluations on promising or informative candidates and refines the skill population based on execution feedback. In experiments across six agent benchmarks and three target models, COBRA‑Skills outperforms existing methods, cuts optimization cost by 55–58 % compared to SkillOpt, and requires only 50 unique optimization examples per benchmark.
By Pingchen Lu, Xiangyi Wang, Xiang Li, Jie Mao, Zikun Qu, Junfeng Luo, Yao Shu, Bryan Kian Hsiang Low, Zhongxiang Dai
arXiv:2608. 03437v1 Announce Type: cross Abstract: While human evaluation is the gold standard in many NLP tasks, it suffers from prohibitive costs and poor scalability.
By Vil\'em Zouhar, Julia Kreutzer, Alon Lavie, Tom Kocmi, Matt Post, Ond\v{r}ej Bojar, Mrinmaya Sachan
Goldilocks RL proposes an adaptive data‑selection strategy for reinforcement learning in language models, using a Selector network to predict reward variability and prioritize questions that are neither too easy nor too hard. By continuously adapting to the model’s evolving abilities, Goldilocks improves training efficiency on large‑scale reasoning datasets, achieving up to 78% fewer optimization steps compared to standard GRPO. The approach addresses the sample‑inefficiency problem of sparse rewards in reasoning tasks.
By Ilia Mahrooghi, Aryo Lotfi, Emmanuel Abbe
The paper introduces Poodle, a prototype for just‑in‑time model replacement (JITR) that automatically swaps a large language model with a cheaper, task‑specific model when a recurring task is detected. Poodle reduces inference time by up to 7.5× and saves over $2,200 per 1 M requests compared to a flagship hosted LLM, while maintaining competitive accuracy. The authors argue that model search and transfer learning are essential for efficiently identifying and fine‑tuning these custom models.
By Nils Strassenburg, Boris Glavic, Tilmann Rabl
The paper introduces DUCB-OGD, an algorithm that couples a Discounted Upper‑Confidence‑Bound sampler with Online Gradient Descent to address dynamic minimax regret in robust large‑language‑model post‑training. It operates under instantaneous mini‑batch‑only bandit feedback, tracking worst‑source performance without re‑evaluating historical data. Experiments on fine‑tuning, preference optimization, and reinforcement learning demonstrate that DUCB‑OGD improves worst‑group robustness with negligible computational overhead.