让数字主播快速适应动态任务配置，实现低延迟高精度直播互动。
Training Agents to Evolve with Their Harness: TaoLive Digital Avatar Agent Technical Report
- 通过感知任务状态的增强训练，使小模型适应不断变化的任务配置。
- 在直播问答和配置变异测试中分别达到94.8和94.6分，显著优于基线。
- 可在单张NVIDIA H20 GPU上运行，适合实际直播场景部署。
AI驱动的数字主播需实时回答商品问题、与观众互动并执行营销策略，对低延迟、频繁策略更新及准确有效响应提出高要求。可演化编排（Evolvable Harnesses）允许独立更新技能、钩子、提示和工具，但带来权衡：大模型零样本适应但速度慢，小模型满足延迟要求却易过拟合固定编排。本文提出面向编排的训练（HAT），训练紧凑模型以适应变化的编排。其核心为编排状态增强（HSA），对技能标识符、内容、工具模式、提示结构和钩子函数施加任务保持变换。训练分三阶段：HSA-SFT从强模型轨迹中学习推理与工具使用；通用在线策略蒸馏恢复微调中丢失的泛化能力；HSA-RL在增强环境中通过强化学习提升对编排变化的鲁棒性。在四个评估集上，HAT在直播问答任务中得分94.8（基线80.3，最强通用大模型93.0），在编排变异问答中达94.6（基线75.4）。相比固定编排微调，后者使IFEval下降7.7分，而HAT避免性能下降，达到83.5。系统在单张NVIDIA H20 GPU上实现P50延迟3.4秒、P95延迟8.1秒。部署于淘宝直播数字主播服务后，线上A/B测试显示GMV和商品页浏览量均有正向提升。
原文摘要 · Abstract (English)
AI-powered digital avatar streamers must answer product questions, engage viewers, and execute marketing strategies in real time, demanding low latency, frequent strategy updates, and accurate yet effective responses. Evolvable Harnesses, whose Skills, Hooks, prompts, and tools can be updated independently of model weights, enable rapid iteration but expose a trade-off: large models adapt zero-shot yet are too slow, whereas compact models meet latency targets but overfit to fixed Harness configurations. We propose Harness-Aware Training (HAT), which trains compact models to adapt to changing Harnesses. Its key component, Harness-State Augmentation (HSA), applies task-preserving transformations to Skill identifiers and content, tool schemas, prompt structures, and Hook functions. Training proceeds in three stages: HSA-SFT learns reasoning and tool use from strong-model trajectories across diverse environments; General On-Policy Distillation restores generalization lost during SFT; and HSA-RL improves robustness to changing Harnesses through reinforcement learning in augmented environments. Across four evaluation sets, HAT achieves 94.8 on Live-Stream QA (base: 80.3; strongest general LLM: 93.0) and 94.6 on Harness-Variant QA (base: 75.4). Unlike Fixed-Harness SFT, which lowers IFEval by 7.7 points from the base model, HAT avoids this regression and reaches 83.5. On one NVIDIA H20 GPU, the optimized system delivers P50 and P95 latencies of 3.4 s and 8.1 s. Deployed in Taobao Live's digital-avatar service, it also yields positive online A/B test results for GMV and item-page views.
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。