Shopify 在 PyTorchCon 2026 分享基于 PyTorch 和 vLLM 的持续学习闭环
Shopify 现场拆了自家 Buyer Profile 的持续学习闭环：judge 挑坏输出，critic 诊断，修复回灌训练集，小模型还能打赢 frontier 模型。
Shopify 工程师 Cody Mazza-Anthony 在 PyTorchCon 2026 上讲解 Buyer Profile 场景的持续学习流程。生产环境中由校准过的 judge 模型标记低分输出，critic 对每次失败做诊断，再带 critic 提示进行 replay 并由 judge 重新打分。被接受的修复会扩充训练数据，形成数据飞轮。演讲还展示了一个小型专用模型在该任务上跑赢 frontier 模型，并通过 prompt 压缩优化推理成本。
Now: @Shopify’s continual learning loop with PyTorch and @vLLM_project takes the stage at #PyTorchCon 2026 with Cody Mazza-Anthony @cmazzaanthony.
Using Buyer Profile as an example, Cody walks through how a calibrated judge flags low-scoring outputs in production. Critics diagnose each failure, and a replay with a critic hint is scored again by the judge. Accepted repairs expand and improve the training data.
Cody shows how a small specialist model can outperform a frontier model on this task, then turns to optimizing inference through prompt compression.
Watch the keynote livestream: https://t.co/6TtaEkw0x7