评测与研究

typesafe-local

@aabolfazl★ 6PythonMIT更新于 2026-09-18▶ 可当场跑

受TypeSafe Ai启发，向本地LLM提问，获得校准概率而非文本，输出结构化结果，无需生成或解析，支持MLX和Apple Silicon。

英文原文

Inspired by TypeSafe Ai, Ask a local LLM typed questions, get calibrated probabilities instead of text. Structured output without generation or parsing. MLX / Apple Silicon.

aabolfazl/typesafe-local ↗

它在哪儿调用了 Jev

Then: curl -s localhost:8000/v1/systemone -H 'content-type: application/json' -d @examples/support_ticket.json

ots/server.py:8

链接指向我们抓取当天的那个 commit，行号是准的。

它问 Jev 的问题

下面是从这个项目源码里原样取出来的 question 组合。

取自 tests/test_with_model.py:80

  1. a是/否

    Does the customer request a refund?

  2. b是/否

    Is the customer angry?

用你自己的内容跑一遍

代码
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    a: { type: 'boolean', instructions: 'Does the customer request a refund?' },
    b: { type: 'boolean', instructions: 'Is the customer angry?' },
  },
});

同类的其他项目