Search papers, labs, and topics across Lattice.
This paper introduces Triton, a dataset and progressive training curriculum to improve the robustness of text-based web agents. The curriculum uses Structural-Semantic Hard Negative Mining and a Dual-Agent Consensus pipeline to create training data, then trains models with SFT, Odds Ratio Preference Optimization (ORPO), and Group Relative Policy Optimization (GRPO). Triton-GRPO-32B achieves a new state-of-the-art step success rate of 58.7% on Mind2Web, outperforming GPT-4.5 and Claude-4.5, demonstrating the effectiveness of specialized data curriculum over model scale.
Forget scaling laws: a carefully crafted data curriculum lets a 32B parameter model beat GPT-4.5 and Claude-4.5 on web navigation.
Text-based web agents offer computational efficiency for autonomous web navigation, yet developing robust agents remains challenging due to the noisy and heterogeneous nature of real-world HTML. Standard Supervised Fine-Tuning (SFT) approaches fail in two critical dimensions: they lack discrimination capabilities to reject plausible but incorrect elements in densely populated pages, and exhibit limited generalization to unseen website layouts. To address these challenges, we introduce the Triton dataset (590k instances) and a progressive training curriculum. Triton is constructed via Structural-Semantic Hard Negative Mining, which explicitly mines topologically similar distractors, and a Dual-Agent Consensus pipeline that synthesizes diverse cross-domain tasks with strict verification. Building upon this foundation, our progressive curriculum produces three models: Triton-SFT-32B for basic imitation, Triton-ORPO-32B for robust discrimination via Odds Ratio Preference Optimization, and Triton-GRPO-32B for long-horizon consistency through Group Relative Policy Optimization. Empirical evaluation on Mind2Web demonstrates that Triton-GRPO-32B achieves state-of-the-art performance among open-source models with 58.7% Step Success Rate, surpassing GPT-4.5 (42.4%) and Claude-4.5 (41.4%) by over 16%, validating that specialized data curriculum outweighs raw parameter scale for web navigation.
