| 1 | import { describe, expect, it } from "@effect/vitest"; |
| 2 | import { MuseSettings } from "../settings.ts"; |
| 3 | import * as Effect from "effect/Effect"; |
| 4 | import * as Schema from "effect/Schema"; |
| 5 | |
| 6 | import type { MuseSdkHost } from "./sdk.ts"; |
| 7 | import { |
| 8 | discoverMuseModels, |
| 9 | museModelCapabilities, |
| 10 | resolveMuseReasoningEffort, |
| 11 | } from "./modelCatalog.ts"; |
| 12 | |
| 13 | const settings = Schema.decodeSync(MuseSettings)({}); |
| 14 | const host = (catalog: Record<string, unknown>): MuseSdkHost => ({ |
| 15 | connection: { |
| 16 | request: async () => catalog, |
| 17 | command: async () => ({}), |
| 18 | mintCommandId: () => "test", |
| 19 | onNotification: () => {}, |
| 20 | onServerRequest: () => {}, |
| 21 | onProtocolError: () => {}, |
| 22 | closed: new Promise<void>(() => {}), |
| 23 | }, |
| 24 | initializeResult: { grantedCapabilities: [] }, |
| 25 | exited: new Promise(() => {}), |
| 26 | close: async () => {}, |
| 27 | }); |
| 28 | |
| 29 | const row = (modelId: string, extra: Record<string, unknown> = {}) => ({ |
| 30 | modelId, |
| 31 | displayLabel: modelId, |
| 32 | providerId: "meta", |
| 33 | profileId: "tbh", |
| 34 | isDefault: false, |
| 35 | contextLimit: 1_000_000, |
| 36 | ...extra, |
| 37 | }); |
| 38 | |
| 39 | describe("Muse model catalog", () => { |
| 40 | it.effect("uses the labels, efforts and defaults that model/list sends", () => |
| 41 | Effect.gen(function* () { |
| 42 | const models = yield* discoverMuseModels(settings, {}, undefined, async () => |
| 43 | host({ |
| 44 | providerId: "meta", |
| 45 | profileId: "tbh", |
| 46 | source: "providerCatalog", |
| 47 | models: [ |
| 48 | row("restful-walrus", { |
| 49 | displayLabel: " Restful Walrus ", |
| 50 | isDefault: true, |
| 51 | variants: ["minimal", "low", "high", "future-tier"], |
| 52 | reasoningEffortVariants: [{ tier: "low", description: "Quick" }], |
| 53 | defaultReasoningEffort: "high", |
| 54 | }), |
| 55 | row("older-host"), |
| 56 | row("unknown-efforts", { variants: "unknown" }), |
| 57 | row("foreign", { providerId: "another" }), |
| 58 | ], |
| 59 | }), |
| 60 | ).pipe(Effect.scoped); |
| 61 | expect(models.map(({ slug, name, isDefault }) => [slug, name, isDefault])).toEqual([ |
| 62 | ["restful-walrus", "Restful Walrus", true], |
| 63 | ["older-host", "older-host", false], |
| 64 | ["unknown-efforts", "unknown-efforts", false], |
| 65 | ]); |
| 66 | expect(models[0]?.capabilities?.optionDescriptors).toEqual([ |
| 67 | { |
| 68 | id: "reasoningEffort", |
| 69 | label: "Reasoning", |
| 70 | type: "select", |
| 71 | currentValue: "high", |
| 72 | options: [ |
| 73 | { id: "minimal", label: "Minimal" }, |
| 74 | { id: "low", label: "Low", description: "Quick" }, |
| 75 | { id: "high", label: "High", isDefault: true }, |
| 76 | ], |
| 77 | }, |
| 78 | ]); |
| 79 | // Older hosts and "unknown" fall back to Muse's documented tiers. |
| 80 | expect(models[1]?.capabilities).toEqual(museModelCapabilities()); |
| 81 | expect(models[2]?.capabilities).toEqual(museModelCapabilities()); |
| 82 | }), |
| 83 | ); |
| 84 | |
| 85 | it("normalizes saved and implicit efforts against the selected model before dispatch", () => { |
| 86 | const fallback = museModelCapabilities(); |
| 87 | expect(resolveMuseReasoningEffort(fallback, undefined)).toBe("high"); |
| 88 | expect(resolveMuseReasoningEffort(fallback, "max")).toBe("max"); |
| 89 | expect(resolveMuseReasoningEffort(fallback, "ultra")).toBe("high"); |
| 90 | const restricted = museModelCapabilities({ variants: ["xhigh"] }); |
| 91 | expect(resolveMuseReasoningEffort(restricted, undefined)).toBe("xhigh"); |
| 92 | expect(resolveMuseReasoningEffort(restricted, "low")).toBe("xhigh"); |
| 93 | expect( |
| 94 | resolveMuseReasoningEffort(museModelCapabilities({ variants: [] }), "max"), |
| 95 | ).toBeUndefined(); |
| 96 | expect(resolveMuseReasoningEffort(undefined, "max")).toBe("max"); |
| 97 | }); |
| 98 | }); |