mirror of
https://github.com/hanzoai/extension.git
synced 2026-10-31 21:57:06 +00:00
zeekay
69fdebf07d
The overlay hard-coded five models, three of them third-party and none in /v1/models. It now lists Auto and the Zen chat rows of chat.listModels, and resolves Auto the way the sidebar does, where Auto now means zen5-flash for a short prompt and zen5 otherwise.
69 lines
2.4 KiB
TypeScript
69 lines
2.4 KiB
TypeScript
import { describe, expect, it } from 'vitest';
|
|
import {
|
|
InFlightRequestDeduper,
|
|
makeChatRequestKey,
|
|
pickModelForTokenBudget,
|
|
trimMessagesToBudget,
|
|
} from '../src/runtime-guard';
|
|
|
|
describe('runtime-guard', () => {
|
|
it('trims older messages to respect token budget while keeping newest', () => {
|
|
const messages = [
|
|
{ role: 'system', content: 'sys '.repeat(200) },
|
|
{ role: 'user', content: 'older '.repeat(400) },
|
|
{ role: 'assistant', content: 'older answer '.repeat(300) },
|
|
{ role: 'user', content: 'latest prompt' },
|
|
];
|
|
|
|
const result = trimMessagesToBudget(messages as any, {
|
|
maxMessages: 24,
|
|
maxInputTokens: 300,
|
|
});
|
|
|
|
expect(result.messages.length).toBeGreaterThan(0);
|
|
expect(result.messages[result.messages.length - 1].content).toContain('latest prompt');
|
|
expect(result.droppedMessages).toBeGreaterThan(0);
|
|
expect(result.inputTokens).toBeLessThanOrEqual(300 + 10);
|
|
});
|
|
|
|
it('creates deterministic chat request keys', () => {
|
|
const messages = [{ role: 'user', content: 'hello world' }];
|
|
const k1 = makeChatRequestKey('gpt-4o', messages as any, 0.2, 500);
|
|
const k2 = makeChatRequestKey('gpt-4o', messages as any, 0.2, 500);
|
|
const k3 = makeChatRequestKey('gpt-4o', [{ role: 'user', content: 'hello world!' }] as any, 0.2, 500);
|
|
|
|
expect(k1).toBe(k2);
|
|
expect(k1).not.toBe(k3);
|
|
});
|
|
|
|
it('dedupes in-flight requests with the same key', async () => {
|
|
const deduper = new InFlightRequestDeduper<string>();
|
|
let runs = 0;
|
|
|
|
const first = deduper.getOrCreate('same', async () => {
|
|
runs += 1;
|
|
await new Promise((resolve) => setTimeout(resolve, 25));
|
|
return 'ok';
|
|
});
|
|
const second = deduper.getOrCreate('same', async () => {
|
|
runs += 1;
|
|
return 'second';
|
|
});
|
|
|
|
const [a, b] = await Promise.all([first.promise, second.promise]);
|
|
|
|
expect(a).toBe('ok');
|
|
expect(b).toBe('ok');
|
|
expect(first.deduped).toBe(false);
|
|
expect(second.deduped).toBe(true);
|
|
expect(runs).toBe(1);
|
|
});
|
|
|
|
it('routes auto to zen5-flash for a short prompt and zen5 otherwise', () => {
|
|
expect(pickModelForTokenBudget('auto', ['zen5', 'zen5-flash'], 200)).toBe('zen5-flash');
|
|
expect(pickModelForTokenBudget('auto', ['zen5', 'zen5-flash'], 5000)).toBe('zen5');
|
|
expect(pickModelForTokenBudget('auto', [], 200)).toBe('zen5');
|
|
expect(pickModelForTokenBudget('zen6', ['zen5'], 200)).toBe('zen6');
|
|
});
|
|
});
|