通过逆向生成任务提升大模型工具使用泛化能力
DIVE: Scaling Diversity in Agentic Task Synthesis for Generalizable Tool Use
- 先执行真实工具再反推任务，确保任务可执行且有真实依据
- 在5个领域373个工具上构建多样化任务，使模型在9个外部测试集平均提升22分
- 证明多样性比数据量更能提升泛化，4倍少数据仍更优
近期工作尝试为后训练的工具使用大模型合成智能体任务，但面对任务与工具集变化时的鲁棒泛化仍是未解难题。我们发现其脆弱性源于合成任务多样性不足。提升多样性困难在于：训练要求任务可执行、可验证，而泛化又需覆盖多样工具类型、组合方式及异构使用模式。为此提出DIVE，一种证据驱动的合成方法：先执行多样化的真实世界工具，再严格反推由执行轨迹所蕴含的任务，从源头保证任务的可执行性。DIVE沿两个可控维度扩展结构多样性——工具池覆盖率和单任务工具组合丰富度，并通过‘证据收集—任务推导’循环，在五个领域中实现373个工具上的多步工具使用模式。在48,000条SFT数据与3,200条RL数据上训练Qwen3-8B模型，使其在9个分布外（OOD）基准上平均提升22分，优于最强的8B基线模型68分。令人惊讶的是，控制变量分析表明，多样性扩展始终优于单纯增加数据量，即使数据量仅为四分之一也表现更优。
原文摘要 · Abstract (English)
Recent work synthesizes agentic tasks for post-training tool-using LLMs, yet robust generalization under shifts in tasks and toolsets remains an open challenge. We trace this brittleness to insufficient diversity in synthesized tasks. Scaling diversity is difficult because training requires tasks to remain executable and verifiable, while generalization demands coverage of diverse tool types, toolset combinations, and heterogeneous tool-use patterns. We propose DIVE, an evidence-driven recipe that inverts synthesis order, executing diverse, real-world tools first and reverse-deriving tasks strictly entailed by the resulting traces, thereby providing grounding by construction. DIVE scales structural diversity along two controllable axes, tool-pool coverage and per-task toolset variety, and an Evidence Collection--Task Derivation loop further induces rich multi-step tool-use patterns across 373 tools in five domains. Training Qwen3-8B on DIVE data (48k SFT + 3.2k RL) improves by +22 average points across 9 OOD benchmarks and outperforms the strongest 8B baseline by +68. Remarkably, controlled scaling analysis reveals that diversity scaling consistently outperforms quantity scaling for OOD generalization, even with 4x less data.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。