feat: 采集 LLM token 用量并上报 lpt-be
- 客户端解析上游 usage(含 DeepSeek 缓存命中/未命中),跨重试累加 - 新增计价表(三档单价 + 峰谷时段),费用在结算时冻结 - 新增用量上报与本地 outbox 重试;4xx 判定为确定性拒绝,不无限重试 - 面板补充 Token/费用/调用次数与计价快照 - 顺带修复面板内联脚本转义错误(模板字面量吃掉 \n 与 \',面板一直空白)
This commit is contained in:
@@ -0,0 +1,134 @@
|
||||
import { describe, expect, it } from "vitest";
|
||||
|
||||
import {
|
||||
accumulateUsage,
|
||||
combineSource,
|
||||
emptyUsage,
|
||||
hasUsage,
|
||||
normalizeUsage,
|
||||
type TokenUsage,
|
||||
} from "./usage.js";
|
||||
|
||||
function usage(partial: Partial<TokenUsage>): TokenUsage {
|
||||
return { ...emptyUsage(), ...partial };
|
||||
}
|
||||
|
||||
describe("normalizeUsage", () => {
|
||||
it("取 DeepSeek 顶层字段,包括缓存命中与未命中", () => {
|
||||
const { usage: result, source } = normalizeUsage({
|
||||
prompt_tokens: 1000,
|
||||
completion_tokens: 200,
|
||||
total_tokens: 1200,
|
||||
prompt_cache_hit_tokens: 800,
|
||||
prompt_cache_miss_tokens: 200,
|
||||
completion_tokens_details: { reasoning_tokens: 50 },
|
||||
});
|
||||
|
||||
expect(source).toBe("upstream");
|
||||
expect(result).toEqual({
|
||||
promptTokens: 1000,
|
||||
completionTokens: 200,
|
||||
totalTokens: 1200,
|
||||
cacheHitTokens: 800,
|
||||
cacheMissTokens: 200,
|
||||
reasoningTokens: 50,
|
||||
});
|
||||
});
|
||||
|
||||
it("OpenAI 风格的 prompt_tokens_details.cached_tokens 也能识别", () => {
|
||||
const { usage: result } = normalizeUsage({
|
||||
prompt_tokens: 500,
|
||||
completion_tokens: 100,
|
||||
prompt_tokens_details: { cached_tokens: 300 },
|
||||
});
|
||||
|
||||
expect(result.cacheHitTokens).toBe(300);
|
||||
// 未命中数缺失时用 prompt - hit 兜底推出
|
||||
expect(result.cacheMissTokens).toBe(200);
|
||||
});
|
||||
|
||||
it("total_tokens 缺失时用 prompt + completion 兜底", () => {
|
||||
const { usage: result } = normalizeUsage({ prompt_tokens: 10, completion_tokens: 5 });
|
||||
expect(result.totalTokens).toBe(15);
|
||||
});
|
||||
|
||||
it("缓存命中数大于 prompt 时不会算出负的未命中数", () => {
|
||||
const { usage: result } = normalizeUsage({
|
||||
prompt_tokens: 100,
|
||||
prompt_tokens_details: { cached_tokens: 150 },
|
||||
});
|
||||
expect(result.cacheMissTokens).toBe(0);
|
||||
});
|
||||
|
||||
it("非对象一律当成未上报,不抛错", () => {
|
||||
for (const raw of [undefined, null, "usage", 42, []]) {
|
||||
const { usage: result, source } = normalizeUsage(raw);
|
||||
expect(source).toBe("missing");
|
||||
expect(hasUsage(result)).toBe(false);
|
||||
}
|
||||
});
|
||||
|
||||
it("负值与非数字字段按未上报处理,不猜值", () => {
|
||||
const { usage: result, source } = normalizeUsage({
|
||||
prompt_tokens: -5,
|
||||
completion_tokens: "abc",
|
||||
});
|
||||
expect(source).toBe("missing");
|
||||
expect(result.promptTokens).toBeNull();
|
||||
expect(result.completionTokens).toBeNull();
|
||||
});
|
||||
|
||||
it("明确上报 0 与未上报是两种结果", () => {
|
||||
const reportedZero = normalizeUsage({ prompt_tokens: 0, completion_tokens: 0 });
|
||||
expect(reportedZero.source).toBe("upstream");
|
||||
expect(reportedZero.usage.promptTokens).toBe(0);
|
||||
|
||||
const missing = normalizeUsage({});
|
||||
expect(missing.source).toBe("missing");
|
||||
expect(missing.usage.promptTokens).toBeNull();
|
||||
});
|
||||
});
|
||||
|
||||
describe("accumulateUsage", () => {
|
||||
it("累加各次尝试的用量", () => {
|
||||
const result = accumulateUsage([
|
||||
usage({ promptTokens: 100, completionTokens: 20, totalTokens: 120 }),
|
||||
usage({ promptTokens: 150, completionTokens: 30, totalTokens: 180 }),
|
||||
]);
|
||||
|
||||
expect(result.promptTokens).toBe(250);
|
||||
expect(result.completionTokens).toBe(50);
|
||||
expect(result.totalTokens).toBe(300);
|
||||
});
|
||||
|
||||
it("某次尝试没上报的字段按 0 计,但全都没上报时保持 null", () => {
|
||||
const result = accumulateUsage([
|
||||
usage({ promptTokens: 100 }),
|
||||
usage({}),
|
||||
]);
|
||||
|
||||
expect(result.promptTokens).toBe(100);
|
||||
// 两次都没上报的字段不能被算成 0
|
||||
expect(result.completionTokens).toBeNull();
|
||||
expect(result.reasoningTokens).toBeNull();
|
||||
});
|
||||
|
||||
it("空数组返回全 null", () => {
|
||||
expect(hasUsage(accumulateUsage([]))).toBe(false);
|
||||
});
|
||||
});
|
||||
|
||||
describe("combineSource", () => {
|
||||
it("只要有真实上游数据就标 upstream", () => {
|
||||
expect(combineSource(["missing", "upstream", "missing"])).toBe("upstream");
|
||||
});
|
||||
|
||||
it("没有 upstream 但有估算时标 estimated", () => {
|
||||
expect(combineSource(["missing", "estimated"])).toBe("estimated");
|
||||
});
|
||||
|
||||
it("全为 missing 时标 missing", () => {
|
||||
expect(combineSource(["missing", "missing"])).toBe("missing");
|
||||
expect(combineSource([])).toBe("missing");
|
||||
});
|
||||
});
|
||||
Reference in New Issue
Block a user