SDK 与客户端
将本地语言模型转换为兼容Jev的结构化决策引擎,支持Choice、Score和Noul输出,采用预填充二进制推理。
Adapt local language models into Jev-compatible structured decision engines with Choice, Score, and Noul outputs powered by prefill-only binary inference.
if any(getattr(route, "path", None) == "/v1/systemone" for route in app.routes):src/llm2jev/sglang_server.py:122
链接指向我们抓取当天的那个 commit,行号是准的。
下面是从这个项目源码里原样取出来的 question 组合。
取自 examples/sglang_inference.py:14
delivery是/否这是否是一个物流配送问题?
department选项哪个部门应该处理这个请求?
shipping — 物流配送billing — 付款账单technical — 产品故障urgency打分这个请求有多紧急?
0 — 不紧急1 — 比较紧急2 — 非常紧急import { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
delivery: { type: 'boolean', instructions: '这是否是一个物流配送问题?' },
department: { type: 'choice', instructions: '哪个部门应该处理这个请求?',
criteria: { shipping: '物流配送', billing: '付款账单', technical: '产品故障' } },
urgency: { type: 'score', instructions: '这个请求有多紧急?',
criteria: ['不紧急', '比较紧急', '非常紧急'] },
},
});取自 examples/transformers_inference.py:30
is_delivery_issue是/否这是否是一个物流配送问题?
department选项这个请求应该交给哪个部门处理?
billing — 付款、账单或退款问题shipping — 物流、配送或包裹丢失问题technical — 产品故障或技术支持问题urgency打分评估这个客户请求的紧急程度。
0 — 不紧急1 — 比较紧急2 — 非常紧急import { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
is_delivery_issue: { type: 'boolean', instructions: '这是否是一个物流配送问题?' },
department: { type: 'choice', instructions: '这个请求应该交给哪个部门处理?',
criteria: { billing: '付款、账单或退款问题', shipping: '物流、配送或包裹丢失问题', technical: '产品故障或技术支持问题' } },
urgency: { type: 'score', instructions: '评估这个客户请求的紧急程度。',
criteria: ['不紧急', '比较紧急', '非常紧急'] },
},
});refund是/否Is a refund requested?
import { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
refund: { type: 'boolean', instructions: 'Is a refund requested?' },
},
});