Autoregressive Diffusion World Models for Off-Policy Evaluation of LLM Agents

• URL: http://arxiv.org/abs/2606.05558v1
• Date: Thu, 04 Jun 2026 01:13:24 GMT
• Title: Autoregressive Diffusion World Models for Off-Policy Evaluation of LLM Agents
• Abstract summary: evaluating large language model (LLM) agents in multi-turn interactive environments is expensive and risky.<n>We propose ADWM, an evaluation framework that estimates the performance of a new LLM agent policy purely from pre-collected trajectories.
• Score: 35.37126280179998
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Evaluating large language model (LLM) agents in multi-turn interactive environments is expensive and risky, as it requires online environment interaction. We propose ADWM (Autoregressive Diffusion World Model), an evaluation framework that estimates the performance of a new LLM agent policy purely from pre-collected trajectories. The core idea is to learn a latent diffusion world model that simulates how the environment responds to the evaluation policy, without ever executing it in the real environment. Existing diffusion-based OPE methods guide full trajectories in a single pass by jointly diffusing states and actions, an assumption that breaks down for LLM agents whose actions are discrete text that must be sampled from the policy after observing the environment. Unlike autoregressive world models that suffer from compounding errors, ADWM models each transition as an independent denoising process, enabling reliable step-by-step rollouts where the world model and agent alternate in causal order. Crucially, the LLM agent under evaluation directly guides the diffusion generation at each step via a policy-conditioned score function, ensuring that simulated trajectories accurately reflect its decision-making patterns. Empirically, ADWM achieves accurate value estimates and evaluation reliability across diverse multi-turn agent tasks, demonstrating its promise as a practical framework for offline LLM agent evaluation.

Related papers

• Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution [25.26084224278813]
  Role-Agent harnesses a single Large Language Model (LLM) to function concurrently as both the agent and the environment.<n>Role-Agent consistently improves performance, yielding an average gain of over 4% over strong baselines.
  arXiv  Detail & Related papers  (2026-06-09T14:28:07Z)
• COMAP: Co-Evolving World Models and Agent Policies for LLM Agents [14.918267305899619]
  COMAP is a novel framework that co-evolves textual world models and agent policies through closed-loop interaction.<n>At each decision step, the world model predicts future state feedback for candidate actions, and the agent performs future-aware reflection by estimating the reliability of this feedback.<n>The resulting on-policy trajectories are then used to update the world model via self-distillation, allowing it to better match the agent's evolving interaction distribution.
  arXiv  Detail & Related papers  (2026-06-01T15:21:17Z)
• Learning POMDP World Models from Observations with Language-Model Priors [52.11151769796862]
  Partially-observable Markov decision processes (POMDPs) provide a flexible modeling class for internal world models.<n>We ask whether language-model priors can reduce costly interaction by leveraging prior knowledge.<n>We show that emphPinductor matches the performance and sample efficiency of LLM-based POMDP learning methods.
  arXiv  Detail & Related papers  (2026-05-13T16:18:15Z)
• Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL [0.8546394944138487]
  We formalize text-based world modeling as a steerable transition-dynamics problem decomposed into initial state.<n>We curate 239,403 grounded state-action trajectories spanning nine open-source environments and twelve frontier model families.<n>We introduce a plug-and-play GRPO training framework with deterministic state checks, and perform zero-shot transfer ablations on three OOD environments.
  arXiv  Detail & Related papers  (2026-05-07T00:40:32Z)
• LLM Active Alignment: A Nash Equilibrium Perspective [34.54084293479338]
  We develop a game-theoretic framework for predicting and steering the behavior of large language models.<n>Agents choose actively and strategically which groups to align with, yielding an interpretable and behaviorally substantive policy class.
  arXiv  Detail & Related papers  (2026-02-06T16:26:03Z)
• Reinforcement World Model Learning for LLM-based Agents [60.65003139516272]
  Reinforcement World Model Learning (RWML) is a self-conditioned method that learns action-supervised world models for LLM-based agents.<n>Our method aligns simulated next states produced by the model with realized next states observed from the environment.<n>We evaluate our method on ALFWorld and $2$ Bench and observe significant gains over the base model, despite being entirely self-supervised.
  arXiv  Detail & Related papers  (2026-02-05T16:30:08Z)
• Grounded Test-Time Adaptation for LLM Agents [75.62784644919803]
  Large language model (LLM)-based agents struggle to generalize to novel and complex environments.<n>We propose two strategies for adapting LLM agents by leveraging environment-specific information available during deployment.
  arXiv  Detail & Related papers  (2025-11-06T22:24:35Z)
• cMALC-D: Contextual Multi-Agent LLM-Guided Curriculum Learning with Diversity-Based Context Blending [3.939989712024268]
  Contextual MARL (cMARL) addresses this by parameterizing environments with context variables and training a context-agnostic policy.<n>Existing cMARL methods attempt to use curriculum learning to help train and evaluate context-agnostic policies.<n>We propose Contextual Multi-Agent LLM-Guided Curriculum Learning with Diversity-Based Context Blending (cMALC-D)
  arXiv  Detail & Related papers  (2025-08-28T14:16:17Z)
• MCPEval: Automatic MCP-based Deep Evaluation for AI Agent Models [76.72220653705679]
  We introduce MCPEval, an open-source framework that automates end-to-end task generation and deep evaluation of intelligent agents.<n> MCPEval standardizes metrics, seamlessly integrates with native agent tools, and eliminates manual effort in building evaluation pipelines.<n> Empirical results across five real-world domains show its effectiveness in revealing nuanced, domain-specific performance.
  arXiv  Detail & Related papers  (2025-07-17T05:46:27Z)
• Revisiting Multi-Agent World Modeling from a Diffusion-Inspired Perspective [54.77404771454794]
  We develop a flexible and robust world model for Multi-Agent Reinforcement Learning (MARL) using diffusion models.<n>Our method, Diffusion-Inspired Multi-Agent world model (DIMA), achieves state-of-the-art performance across multiple multi-agent control benchmarks.
  arXiv  Detail & Related papers  (2025-05-27T09:11:38Z)
• MLE-Dojo: Interactive Environments for Empowering LLM Agents in Machine Learning Engineering [57.156093929365255]
  Gym-style framework for systematically reinforcement learning, evaluating, and improving autonomous large language model (LLM) agents.<n>MLE-Dojo covers diverse, open-ended MLE tasks carefully curated to reflect realistic engineering scenarios.<n>Its fully executable environment supports comprehensive agent training via both supervised fine-tuning and reinforcement learning.
  arXiv  Detail & Related papers  (2025-05-12T17:35:43Z)
• Scaling Autonomous Agents via Automatic Reward Modeling And Planning [52.39395405893965]
  Large language models (LLMs) have demonstrated remarkable capabilities across a range of tasks.<n>However, they still struggle with problems requiring multi-step decision-making and environmental feedback.<n>We propose a framework that can automatically learn a reward model from the environment without human annotations.
  arXiv  Detail & Related papers  (2025-02-17T18:49:25Z)

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.