Merge pull request #1944 from shlokkhemani/add-embedding-tests

chore: add embedding tests

packages/core/src/tests/embedding.test.ts

@@ -0,0 +1,201 @@
+import { describe, test, expect, vi, beforeEach } from "vitest";
+import {
+    embed,
+    getEmbeddingConfig,
+    getEmbeddingType,
+    getEmbeddingZeroVector,
+} from "../embedding.ts";
+import { IAgentRuntime, ModelProviderName } from "../types.ts";
+import settings from "../settings.ts";
+
+// Mock environment-related settings
+vi.mock("../settings", () => ({
+    default: {
+        USE_OPENAI_EMBEDDING: "false",
+        USE_OLLAMA_EMBEDDING: "false",
+        USE_GAIANET_EMBEDDING: "false",
+        OPENAI_API_KEY: "mock-openai-key",
+        OPENAI_API_URL: "https://api.openai.com/v1",
+        GAIANET_API_KEY: "mock-gaianet-key",
+        OLLAMA_EMBEDDING_MODEL: "mxbai-embed-large",
+        GAIANET_EMBEDDING_MODEL: "nomic-embed",
+    },
+}));
+
+// Mock fastembed module for local embeddings
+vi.mock("fastembed", () => ({
+    FlagEmbedding: {
+        init: vi.fn().mockResolvedValue({
+            queryEmbed: vi
+                .fn()
+                .mockResolvedValue(new Float32Array(384).fill(0.1)),
+        }),
+    },
+    EmbeddingModel: {
+        BGESmallENV15: "BGE-small-en-v1.5",
+    },
+}));
+
+// Mock global fetch for remote embedding requests
+const mockFetch = vi.fn();
+(global as any).fetch = mockFetch;
+
+describe("Embedding Module", () => {
+    let mockRuntime: IAgentRuntime;
+
+    beforeEach(() => {
+        // Prepare a mock runtime
+        mockRuntime = {
+            character: {
+                modelProvider: ModelProviderName.OLLAMA,
+                modelEndpointOverride: null,
+            },
+            token: "mock-token",
+            messageManager: {
+                getCachedEmbeddings: vi.fn().mockResolvedValue([]),
+            },
+        } as unknown as IAgentRuntime;
+
+        vi.clearAllMocks();
+        mockFetch.mockReset();
+    });
+
+    describe("getEmbeddingConfig", () => {
+        test("should return BGE config by default", () => {
+            const config = getEmbeddingConfig();
+            expect(config.dimensions).toBe(384);
+            expect(config.model).toBe("BGE-small-en-v1.5");
+            expect(config.provider).toBe("BGE");
+        });
+
+        test("should return OpenAI config when USE_OPENAI_EMBEDDING is true", () => {
+            vi.mocked(settings).USE_OPENAI_EMBEDDING = "true";
+            const config = getEmbeddingConfig();
+            expect(config.dimensions).toBe(1536);
+            expect(config.model).toBe("text-embedding-3-small");
+            expect(config.provider).toBe("OpenAI");
+        });
+    });
+
+    describe("getEmbeddingType", () => {
+        test("should return 'remote' for Ollama provider", () => {
+            const type = getEmbeddingType(mockRuntime);
+            expect(type).toBe("remote");
+        });
+
+        test("should return 'remote' for OpenAI provider", () => {
+            mockRuntime.character.modelProvider = ModelProviderName.OPENAI;
+            const type = getEmbeddingType(mockRuntime);
+            expect(type).toBe("remote");
+        });
+    });
+
+    describe("getEmbeddingZeroVector", () => {
+        beforeEach(() => {
+            vi.mocked(settings).USE_OPENAI_EMBEDDING = "false";
+            vi.mocked(settings).USE_OLLAMA_EMBEDDING = "false";
+            vi.mocked(settings).USE_GAIANET_EMBEDDING = "false";
+        });
+
+        test("should return 384-length zero vector by default (BGE)", () => {
+            const vector = getEmbeddingZeroVector();
+            expect(vector).toHaveLength(384);
+            expect(vector.every((val) => val === 0)).toBe(true);
+        });
+
+        test("should return 1536-length zero vector for OpenAI if enabled", () => {
+            vi.mocked(settings).USE_OPENAI_EMBEDDING = "true";
+            const vector = getEmbeddingZeroVector();
+            expect(vector).toHaveLength(1536);
+            expect(vector.every((val) => val === 0)).toBe(true);
+        });
+    });
+
+    describe("embed function", () => {
+        beforeEach(() => {
+            // Mock a successful remote response with an example 384-dim embedding
+            mockFetch.mockResolvedValue({
+                ok: true,
+                json: () =>
+                    Promise.resolve({
+                        data: [{ embedding: new Array(384).fill(0.1) }],
+                    }),
+            });
+        });
+
+        test("should return an empty array for empty input text", async () => {
+            const result = await embed(mockRuntime, "");
+            expect(result).toEqual([]);
+        });
+
+        test("should return cached embedding if it already exists", async () => {
+            const cachedEmbedding = new Array(384).fill(0.5);
+            mockRuntime.messageManager.getCachedEmbeddings = vi
+                .fn()
+                .mockResolvedValue([{ embedding: cachedEmbedding }]);
+
+            const result = await embed(mockRuntime, "test input");
+            expect(result).toBe(cachedEmbedding);
+        });
+
+        test("should handle local embedding successfully (fastembed fallback)", async () => {
+            // By default, it tries local first if in Node.
+            // Then uses the mock fastembed response above.
+            const result = await embed(mockRuntime, "test input");
+            expect(result).toHaveLength(384);
+            expect(result.every((v) => typeof v === "number")).toBe(true);
+        });
+
+        test("should fallback to remote if local embedding fails", async () => {
+            // Force fastembed import to fail
+            vi.mock("fastembed", () => {
+                throw new Error("Module not found");
+            });
+
+            // Mock a valid remote response
+            const mockResponse = {
+                ok: true,
+                json: () =>
+                    Promise.resolve({
+                        data: [{ embedding: new Array(384).fill(0.1) }],
+                    }),
+            };
+            mockFetch.mockResolvedValueOnce(mockResponse);
+
+            const result = await embed(mockRuntime, "test input");
+            expect(result).toHaveLength(384);
+            expect(mockFetch).toHaveBeenCalled();
+        });
+
+        test("should throw on remote embedding if fetch fails", async () => {
+            mockFetch.mockRejectedValueOnce(new Error("API Error"));
+            vi.mocked(settings).USE_OPENAI_EMBEDDING = "true"; // Force remote
+
+            await expect(embed(mockRuntime, "test input")).rejects.toThrow(
+                "API Error"
+            );
+        });
+
+        test("should throw on non-200 remote response", async () => {
+            const errorResponse = {
+                ok: false,
+                status: 400,
+                statusText: "Bad Request",
+                text: () => Promise.resolve("Invalid input"),
+            };
+            mockFetch.mockResolvedValueOnce(errorResponse);
+            vi.mocked(settings).USE_OPENAI_EMBEDDING = "true"; // Force remote
+
+            await expect(embed(mockRuntime, "test input")).rejects.toThrow(
+                "Embedding API Error"
+            );
+        });
+
+        test("should handle concurrent embedding requests", async () => {
+            const promises = Array(5)
+                .fill(null)
+                .map(() => embed(mockRuntime, "concurrent test"));
+            await expect(Promise.all(promises)).resolves.toBeDefined();
+        });
+    });
+});