import test from 'node:test'; import assert from 'node:assert/strict'; import { fixture, compatibleFixture, until } from './helpers.mjs'; import { Provider } from '../src/server/provider.mjs'; import { Runtime } from '../src/server/runtime.mjs'; import { completionBody, requestVariants } from '../src/shared/contracts/model-request.mjs'; import { messagesOutputSetting } from '../src/shared/contracts/model-capacity.mjs'; const config = { mode: 'compatible', protocol: 'anthropic-messages', model: 'owned-claude', executionMode: 'unrestricted', capacityMode: 'passthrough', maxOutputTokens: 16384 }; test('Messages required output uses explicit native value, matching catalog maximum or its own connection declaration', () => { const input = [{ role: 'user', content: 'Complete original input' }]; const catalog = { id: config.model, maxOutputTokens: 128000 }; assert.equal(completionBody({ ...config, upstreamModel: catalog }, input, []).max_tokens, 128000); assert.equal(completionBody({ ...config, upstreamModel: catalog, nativeParameters: { max_tokens: 234567 } }, input, []).max_tokens, 234567); assert.equal(completionBody({ ...config, upstreamModel: { ...catalog, id: 'different' } }, input, []).max_tokens, 16384); const routing = { members: [ { id: 'a', model: 'a', protocol: 'anthropic-messages', maxOutputTokens: 32768 }, { id: 'b', model: 'b', protocol: 'anthropic-messages', maxOutputTokens: 65536, upstreamModel: { id: 'b', maxOutputTokens: 256000 } }, { id: 'c', model: 'c', protocol: 'chat-completions', maxOutputTokens: 2048 }, ] }; const variants = requestVariants(completionBody({ ...config, modelRouting: routing }, input, [])); assert.deepEqual(variants.map(body => body.max_tokens), [32768, 256000, undefined]); assert.equal(messagesOutputSetting({ ...config, upstreamModel: catalog }).source, 'upstream_catalog'); }); test('actual passthrough Messages HTTP includes required max_tokens and preserves a different upstream 400 with no retry', { timeout: 30000 }, async t => { let fail = false; const upstream = await compatibleFixture(t, (_url, body) => { assert.equal(body.max_tokens, 98304); assert.ok(body.messages.some(row => row.content.some(block => block.text?.includes('ORIGINAL-END')))); if (fail) return { status: 400, body: { type: 'error', error: { type: 'upstream_error', message: 'owned upstream rejected another option' } } }; return { id: 'msg_owned', type: 'message', role: 'assistant', model: config.model, content: [{ type: 'text', text: 'Owned complete response' }], stop_reason: 'end_turn', usage: { input_tokens: 20, output_tokens: 5 } }; }); const f = fixture(t, { modernDefaults: true }), provider = new Provider(), hub = provider.modelHub(f.store); const connection = hub.upsert({ name: 'Owned Messages', ...config, baseUrl: upstream.baseUrl, requiresKey: false, maxOutputTokens: 98304 }); hub.select({ type: 'connection', id: connection.id }); const runtime = new Runtime(f.store, provider); f.onCleanup(() => runtime.stopAll()); runtime.start(f.session.id, 'ORIGINAL-END'); await until(() => !runtime.isActive(f.session.id), 20000); assert.equal(f.session.status, 'completed', JSON.stringify(f.session.failure)); assert.equal(upstream.requests.length, 1); fail = true; runtime.start(f.session.id, 'ORIGINAL-END failure'); await until(() => !runtime.isActive(f.session.id), 20000); assert.equal(upstream.requests.length, 2); assert.equal(f.session.status, 'failed'); assert.match(JSON.stringify(f.session.failure), /owned upstream rejected another option/); assert.match(JSON.stringify(f.session.failure), /400/); });