How to Train Your LLM Web Agent: A Statistical Diagnosis

• URL: http://arxiv.org/abs/2507.04103v3
• Date: Sun, 02 Nov 2025 19:03:02 GMT
• Title: How to Train Your LLM Web Agent: A Statistical Diagnosis
• Authors: Dheeraj Vattikonda, Santhoshi Ravichandran, Emiliano Penaloza, Hadi Nekoei, Megh Thakkar, Thibault Le Sellier de Chezelles, Nicolas Gontier, Miguel Muñoz-Mármol, Sahar Omidi Shayegan, Stefania Raimondo, Xue Liu, Alexandre Drouin, Laurent Charlin, Alexandre Piché, Alexandre Lacoste, Massimo Caccia,
• Abstract summary: We present the first statistically grounded study on compute allocation for LLM web-agent post-training.<n>Our approach uses a two-stage pipeline, training a Llama 3.1 8B student to imitate a Llama 3.3 70B teacher via supervised fine-tuning (SFT) and on-policy reinforcement learning.<n>Our results show that combining SFT with on-policy RL consistently outperforms either approach alone on both WorkArena and MiniWob++.
• Score: 96.86317871461834
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: LLM-based web agents have recently made significant progress, but much of it has occurred in closed-source systems, widening the gap with open-source alternatives. Progress has been held back by two key challenges: first, a narrow focus on single-step tasks that overlooks the complexity of multi-step web interactions; and second, the high compute costs required to post-train LLM-based web agents. To address this, we present the first statistically grounded study on compute allocation for LLM web-agent post-training. Our approach uses a two-stage pipeline, training a Llama 3.1 8B student to imitate a Llama 3.3 70B teacher via supervised fine-tuning (SFT), followed by on-policy reinforcement learning. We find this process highly sensitive to hyperparameter choices, making exhaustive sweeps impractical. To spare others from expensive trial-and-error, we sample 1,370 configurations and use bootstrapping to estimate effective hyperparameters. Our results show that combining SFT with on-policy RL consistently outperforms either approach alone on both WorkArena and MiniWob++. Further, this strategy requires only 55% of the compute to match the peak performance of pure SFT on MiniWob++, effectively pushing the compute-performance Pareto frontier, and is the only strategy that can close the gap with closed-source models.

Related papers

• DiFFPO: Training Diffusion LLMs to Reason Fast and Furious via Reinforcement Learning [37.20873499361773]
  We propose a unified framework for training masked diffusion large language models (dLLMs) to reason better (furious)<n>We first unify the existing baseline approach by proposing to train surrogate policies via off-policy RL, whose likelihood is much more tractable as an approximation to the true dLLM policy.<n>We also propose a new direction of joint training efficient samplers/controllers of dLLMs policy. Via RL, we incentivize dLLMs' natural multi-token prediction capabilities by letting the model learn to adaptively allocate an inference threshold for each prompt.
  arXiv  Detail & Related papers  (2025-10-02T16:57:24Z)
• Your Reward Function for RL is Your Best PRM for Search: Unifying RL and Search-Based TTS [62.22644307952087]
  We introduce AIRL-S, the first natural unification of RL-based and search-based TTS.<n>We leverage adversarial inverse reinforcement learning (AIRL) combined with group relative policy optimization (GRPO) to learn a dense, dynamic PRM directly from correct reasoning traces.<n>Our results show that our unified approach improves performance by 9 % on average over the base model, matching GPT-4o.
  arXiv  Detail & Related papers  (2025-08-19T23:41:15Z)
• RL-PLUS: Countering Capability Boundary Collapse of LLMs in Reinforcement Learning with Hybrid-policy Optimization [86.30192066451256]
  We propose RL-PLUS, a novel hybrid-policy optimization approach for Large Language Models (LLMs)<n> RL-PLUS synergizes internal exploitation with external data to achieve stronger reasoning capabilities and surpass the boundaries of base models.<n>We provide both theoretical analysis and extensive experiments to demonstrate the superiority and generalizability of our approach.
  arXiv  Detail & Related papers  (2025-07-31T23:55:29Z)
• Multi-Agent Reinforcement Learning for Sample-Efficient Deep Neural Network Mapping [54.65536245955678]
  We present a decentralized multi-agent reinforcement learning (MARL) framework designed to overcome the challenge of sample inefficiency.<n>We introduce an agent clustering algorithm that assigns similar mapping parameters to the same agents based on correlation analysis.<n> Experimental results show our MARL approach improves sample efficiency by 30-300x over standard single-agent RL.
  arXiv  Detail & Related papers  (2025-07-22T05:51:07Z)
• Ring-lite: Scalable Reasoning via C3PO-Stabilized Reinforcement Learning for LLMs [51.21041884010009]
  Ring-lite is a Mixture-of-Experts (MoE)-based large language model optimized via reinforcement learning (RL)<n>Our approach matches the performance of state-of-the-art (SOTA) small-scale reasoning models on challenging benchmarks.
  arXiv  Detail & Related papers  (2025-06-17T17:12:34Z)
• First SFT, Second RL, Third UPT: Continual Improving Multi-Modal LLM Reasoning via Unsupervised Post-Training [37.80193099472551]
  We propose MM-UPT, a simple yet effective framework for unsupervised post-training of MLLMs.<n>Our experiments demonstrate that such training method effectively improves the reasoning ability of Qwen2.5-VL-7B.<n>We extend our framework to a data self-generation setting, designing two strategies that prompt the MLLM to synthesize new training samples.
  arXiv  Detail & Related papers  (2025-05-28T15:11:16Z)
• On the Emergence of Thinking in LLMs I: Searching for the Right Intuition [34.32871896067864]
  We propose a post-training framework called Reinforcement Learning via Self-Play (RLSP)<n> RLSP involves three steps: supervised fine-tuning with human or synthetic demonstrations of the reasoning process, using an exploration reward signal to encourage diverse and efficient reasoning behaviors, and RL training with an outcome verifier to ensure correctness while preventing reward hacking.<n> Empirical studies in the math domain show that RLSP improves reasoning.
  arXiv  Detail & Related papers  (2025-02-10T18:52:04Z)
• Dspy-based Neural-Symbolic Pipeline to Enhance Spatial Reasoning in LLMs [29.735465300269993]
  Large Language Models (LLMs) have demonstrated remarkable capabilities across various tasks, yet they often struggle with spatial reasoning.<n>This paper presents a novel neural-symbolic framework that enhances LLMs' spatial reasoning abilities through iterative feedback between LLMs and Answer Set Programming (ASP)<n>We evaluate our approach on two benchmark datasets: StepGame and SparQA.
  arXiv  Detail & Related papers  (2024-11-27T18:04:05Z)
• Understanding Forgetting in LLM Supervised Fine-Tuning and Preference Learning - A Convex Optimization Perspective [55.66517396157806]
  The widely adopted approach in post-training popular open-source LLMs is to sequentially perform SFT and RLHF/DPO.<n>This is suboptimal in terms of SFT and RLHF/DPO trade-off.<n>We propose a practical joint post-training framework which has theoretical convergence guarantees and empirically outperforms sequential post-training framework.
  arXiv  Detail & Related papers  (2024-10-20T19:38:41Z)
• PLANRL: A Motion Planning and Imitation Learning Framework to Bootstrap Reinforcement Learning [13.564676246832544]
  We introduce PLANRL, a framework that chooses when the robot should use classical motion planning and when it should learn a policy. PLANRL switches between two modes of operation: reaching a waypoint using classical techniques when away from the objects and fine-grained manipulation control when about to interact with objects. We evaluate our approach across multiple challenging simulation environments and real-world tasks, demonstrating superior performance in terms of adaptability, efficiency, and generalization compared to existing methods.
  arXiv  Detail & Related papers  (2024-08-07T19:30:08Z)
• Maximize to Explore: One Objective Function Fusing Estimation, Planning, and Exploration [87.53543137162488]
  We propose an easy-to-implement online reinforcement learning (online RL) framework called textttMEX. textttMEX integrates estimation and planning components while balancing exploration exploitation automatically. It can outperform baselines by a stable margin in various MuJoCo environments with sparse rewards.
  arXiv  Detail & Related papers  (2023-05-29T17:25:26Z)
• MALib: A Parallel Framework for Population-based Multi-agent Reinforcement Learning [61.28547338576706]
  Population-based multi-agent reinforcement learning (PB-MARL) refers to the series of methods nested with reinforcement learning (RL) algorithms. We present MALib, a scalable and efficient computing framework for PB-MARL.
  arXiv  Detail & Related papers  (2021-06-05T03:27:08Z)

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.