Perplexity 决策器
We are releasing Decider v1.3 and Decider Flash v1.3, the successors to our earlier decision models: v1, v1.1, v1.2, and Flash v1.2.
我们发布 Decider v1.3 和 Decider Flash v1.3，它们是早先决策模型 v1、v1.1、v1.2 和 Flash v1.2 的后续版本。
Decider Flash v1.3 is a smaller model distilled from Decider v1.3. Both models are available on Hugging Face under Apache license, through OpenRouter, and the Perplexity Decisions API.
Decider Flash v1.3 是从 Decider v1.3 蒸馏得到的小型模型。两个模型均以 Apache 许可证在 Hugging Face 上提供，也可通过 OpenRouter 和 Perplexity Decisions API 使用。
Perplexity Decider is inspired by Jev, TypeSafe AI’s System One Model. It supports binary and multi-class classification with ordered or unordered labels, returning probabilities that applications can use to classify, route, rank, and evaluate inputs.
Perplexity Decider 的灵感来自 Jev，即 TypeSafe AI 的 System One Model。它支持二分类和多分类，标签可以有序也可以无序，并返回概率，供应用对输入进行分类、路由、排序和评估。
Both v1.3 models improve performance and robustness across a broad range of text and vision tasks through bidirectional attention, better-curated training data, and a refined training recipe. We also introduce locate, a new feature to highlight the selected content for named entity recognition, personal information identification, and extractive question-answering tasks.
两个 v1.3 模型通过双向注意力、更精心整理的训练数据以及改进的训练配方，在广泛的文本和视觉任务上提升了性能与稳健性。我们还引入了 locate，这是一项新功能，可在命名实体识别、个人信息识别和抽取式问答任务中高亮所选内容。
In this post, we explain how we train and evaluate Decider and share how we use it internally at Perplexity.
在本文中，我们说明如何训练和评估 Decider，并分享我们在 Perplexity 内部如何使用它。
What is a decision model?
什么是决策模型？
A decision model is a general-purpose classifier. You give it an input and a set of questions each with its possible answers. For each question, it returns a probability distribution over the answers. It supports 3 question types:
决策模型是一种通用分类器。你给它一个输入以及一组问题，每个问题都带有若干可能答案。对于每个问题，它会返回在各答案上的概率分布。它支持 3 种问题类型：
- Yes / No: binary decisions.
- 是 / 否：二元决策。
- Multiple choice: custom multi-class labels.
- 多项选择：自定义多分类标签。
- Score: ordered levels.
- 评分：有序等级。
Decision models support a wide range of...