CacheRL:Multi-Turn Tool-Calling Agents via Cached Rollouts and Hybrid Reward

• URL: http://arxiv.org/abs/2606.14179v1
• Date: Fri, 12 Jun 2026 07:01:50 GMT
• Title: CacheRL:Multi-Turn Tool-Calling Agents via Cached Rollouts and Hybrid Reward
• Abstract summary: CacheRL is a system for training small agent foundation models.<n>It achieves 92 percent process accuracy on multi-step tool-calling tasks.<n>It approaches GPT-5's 94 percent while requiring 100 times less compute.
• Score: 17.939095034578035
• License: http://creativecommons.org/publicdomain/zero/1.0/
• Abstract: We present CacheRL, a system for training small agent foundation models that achieves 92 percent process accuracy on multi-step tool-calling tasks, approaching GPT-5's 94 percent while requiring 100 times less compute. Our approach addresses three challenges in practical agent training: transferring tool-calling knowledge from large models at scale, enabling reinforcement learning without costly live tool execution, and learning robustly from noisy cached environments. CacheRL introduces three key innovations. First, a hybrid thinking trajectory pipeline augments agent trajectories with LLM-generated reasoning traces, producing training examples that teach models not only what tools to call but also why. Second, the CacheAgentLoop eliminates live execution costs through a three-tier fuzzy cache while preserving trajectory fidelity using token-level masking. Third, a cache-tier-aware reward dynamically adjusts answer-quality weights to avoid penalizing models for cache-induced limitations. Through iterative supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO), CacheRL improves Qwen3-4B-Thinking's validation reward from 0.43 to 0.78. On public agentic tool-calling benchmarks, our model achieves competitive performance against frontier models such as GPT-5. Ablation studies show that removing knowledge transfer reduces performance by 41 percent, while cache-aware rewards contribute a 17 percent improvement. Interestingly, reinforcement learning improves training stability but yields limited gains beyond strong supervised fine-tuning, suggesting that data quality and reward design play a more important role than complex optimization methods in building practical small agent models.

Related papers

• Efficient Agentic Reinforcement Learning with On-Policy Intrinsic Knowledge Boundary Enhancement [16.192937389387982]
  We propose AKBE (Agentic Knowledge Boundary Enhancement), an on-policy method that dynamically probes the model's intrinsic knowledge boundary.<n> Experiments on seven QA benchmarks demonstrate that AKBE improves task accuracy by +1.85 on average and reduces tool calls by 18% over standard agentic RL.
  arXiv  Detail & Related papers  (2026-05-26T12:42:23Z)
• TL-RL-FusionNet: An Adaptive and Efficient Reinforcement Learning-Driven Transfer Learning Framework for Detecting Evolving Ransomware Threats [2.3008238872220548]
  Modern ransomware exhibits polymorphic and evasive behaviors by frequently modifying execution patterns to evade detection.<n>We propose TL-RL-FusionNet, a reinforcement learning (RL)-guided hybrid framework that integrates frozen dual transfer learning (TL) backbone as feature extractors.<n>Experiments show TL-RL-Net achieves 99.1% accuracy, 98.6% precision, 99.6% recall, and 99.74% AUC, outperforming non-RL baselines by up to 2.5% in accuracy and 3.1% in recall.
  arXiv  Detail & Related papers  (2026-04-22T07:08:27Z)
• Training Large Reasoning Models Efficiently via Progressive Thought Encoding [63.254758972725654]
  Large reasoning models (LRMs) excel on complex problems but face a critical barrier to efficiency.<n>We introduce Progressive Thought, a parameter-efficient fine-tuning method that enables LRMs to reason effectively under fixed-size caches.
  arXiv  Detail & Related papers  (2026-02-18T20:03:38Z)
• Just-In-Time Reinforcement Learning: Continual Learning in LLM Agents Without Gradient Updates [53.3717573880076]
  We introduce Just-In-Time Reinforcement Learning (JitRL), a training-free framework that enables test-time policy optimization without any gradient updates.<n>JitRL maintains a dynamic, non-parametric memory of experiences and retrieves relevant trajectories to estimate action advantages on-the-fly.<n>Experiments on WebArena and Jericho demonstrate that JitRL establishes a new state-of-the-art among training-free methods.
  arXiv  Detail & Related papers  (2026-01-26T14:16:51Z)
• Every Step Evolves: Scaling Reinforcement Learning for Trillion-Scale Thinking Model [100.86587937568832]
  Ring-1T is the first open-source, state-of-the-art thinking model with a trillion-scale parameter.<n>It features 1 trillion total parameters and activates approximately 50 billion per token.
  arXiv  Detail & Related papers  (2025-10-21T17:46:14Z)
• Efficient RL-based Cache Vulnerability Exploration by Penalizing Useless Agent Actions [0.0]
  Cache-timing attacks exploit microarchitectural characteristics to leak sensitive data.<n>A method based on Reinforcement Learning (RL) has been proposed to automatically explore vulnerabilities for a given cache structure.<n>We propose a method to identify these useless actions during training and penalize them so that the agent avoids them and the exploration efficiency is improved.
  arXiv  Detail & Related papers  (2025-06-08T15:48:16Z)
• Adaptive $Q$-Network: On-the-fly Target Selection for Deep Reinforcement Learning [18.579378919155864]
  We propose Adaptive $Q$Network (AdaQN) to take into account the non-stationarity of the optimization procedure without requiring additional samples.<n>AdaQN is theoretically sound and empirically validate it in MuJoCo control problems and Atari $2600 games.
  arXiv  Detail & Related papers  (2024-05-25T11:57:43Z)
• Augmenting Unsupervised Reinforcement Learning with Self-Reference [63.68018737038331]
  Humans possess the ability to draw on past experiences explicitly when learning new tasks. We propose the Self-Reference (SR) approach, an add-on module explicitly designed to leverage historical information. Our approach achieves state-of-the-art results in terms of Interquartile Mean (IQM) performance and Optimality Gap reduction on the Unsupervised Reinforcement Learning Benchmark.
  arXiv  Detail & Related papers  (2023-11-16T09:07:34Z)
• Mastering the Unsupervised Reinforcement Learning Benchmark from Pixels [112.63440666617494]
  Reinforcement learning algorithms can succeed but require large amounts of interactions between the agent and the environment. We propose a new method to solve it, using unsupervised model-based RL, for pre-training the agent. We show robust performance on the Real-Word RL benchmark, hinting at resiliency to environment perturbations during adaptation.
  arXiv  Detail & Related papers  (2022-09-24T14:22:29Z)
• Online Convolutional Re-parameterization [51.97831675242173]
  We present online convolutional re- parameterization (OREPA), a two-stage pipeline, aiming to reduce the huge training overhead by squeezing the complex training-time block into a single convolution. Compared with the state-of-the-art re-param models, OREPA is able to save the training-time memory cost by about 70% and accelerate the training speed by around 2x. We also conduct experiments on object detection and semantic segmentation and show consistent improvements on the downstream tasks.
  arXiv  Detail & Related papers  (2022-04-02T09:50:19Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.