smart-coding-mcp 1.4.1 → 2.1.0

package/test/embedding-model.test.js

@@ -9,20 +9,22 @@
  */
 
 import { describe, it, expect, beforeAll } from 'vitest';
-import { pipeline } from '@xenova/transformers';
+import { createEmbedder } from '../lib/mrl-embedder.js';
 import { cosineSimilarity } from '../lib/utils.js';
 import { loadConfig } from '../lib/config.js';
 
 describe('Local Embedding Model', () => {
   let embedder;
   let config;
+  let expectedDimension;
   
   beforeAll(async () => {
     config = await loadConfig();
     console.log(`[Test] Loading embedding model: ${config.embeddingModel}`);
-    embedder = await pipeline('feature-extraction', config.embeddingModel);
-    console.log('[Test] Embedding model loaded successfully');
-  });
+    embedder = await createEmbedder(config);
+    expectedDimension = embedder.dimension;
+    console.log(`[Test] Embedding model loaded (${expectedDimension}d)`);
+  }, 120000);
 
   describe('Model Loading', () => {
     it('should load the embedding model', () => {
@@ -30,8 +32,9 @@ describe('Local Embedding Model', () => {
       expect(typeof embedder).toBe('function');
     });
     
-    it('should use the configured model', () => {
-      expect(config.embeddingModel).toBe('Xenova/all-MiniLM-L6-v2');
+    it('should have model metadata', () => {
+      expect(embedder.modelName).toBeDefined();
+      expect(embedder.dimension).toBeGreaterThan(0);
     });
   });
 
@@ -49,8 +52,8 @@ describe('Local Embedding Model', () => {
       const output = await embedder(text, { pooling: 'mean', normalize: true });
       const vector = Array.from(output.data);
       
-      // MiniLM-L6 produces 384-dimensional vectors
-      expect(vector.length).toBe(384);
+      // Dimension depends on model (MRL: configurable, MiniLM: 384)
+      expect(vector.length).toBe(expectedDimension);
     });
     
     it('should return normalized vectors', async () => {
@@ -85,7 +88,7 @@ describe('Local Embedding Model', () => {
       const output = await embedder(code, { pooling: 'mean', normalize: true });
       const vector = Array.from(output.data);
       
-      expect(vector.length).toBe(384);
+      expect(vector.length).toBe(expectedDimension);
     });
     
     it('should handle multiline text', async () => {
@@ -93,7 +96,7 @@ describe('Local Embedding Model', () => {
       const output = await embedder(multiline, { pooling: 'mean', normalize: true });
       const vector = Array.from(output.data);
       
-      expect(vector.length).toBe(384);
+      expect(vector.length).toBe(expectedDimension);
     });
     
     it('should handle special characters', async () => {
@@ -101,7 +104,7 @@ describe('Local Embedding Model', () => {
       const output = await embedder(special, { pooling: 'mean', normalize: true });
       const vector = Array.from(output.data);
       
-      expect(vector.length).toBe(384);
+      expect(vector.length).toBe(expectedDimension);
     });
   });
 

package/test/helpers.js

@@ -8,7 +8,7 @@ import { EmbeddingsCache } from '../lib/cache.js';
 import { CodebaseIndexer } from '../features/index-codebase.js';
 import { CacheClearer } from '../features/clear-cache.js';
 import { HybridSearch } from '../features/hybrid-search.js';
-import { pipeline } from '@xenova/transformers';
+import { createEmbedder } from '../lib/mrl-embedder.js';
 import fs from 'fs/promises';
 import path from 'path';
 
@@ -22,8 +22,8 @@ let sharedEmbedder = null;
 export async function getEmbedder(config) {
   if (!sharedEmbedder) {
     console.log('[TestHelper] Loading embedding model (first time)...');
-    sharedEmbedder = await pipeline('feature-extraction', config.embeddingModel);
-    console.log('[TestHelper] Embedding model loaded');
+    sharedEmbedder = await createEmbedder(config);
+    console.log(`[TestHelper] Embedding model loaded (${sharedEmbedder.dimension}d)`);
   }
   return sharedEmbedder;
 }

package/test/mrl-embedder.test.js

@@ -0,0 +1,108 @@
+/**
+ * Tests for MRL Embedder
+ * 
+ * Tests the Matryoshka Representation Learning embedder:
+ * - Configurable dimensions (64, 128, 256, 512, 768)
+ * - Layer normalization and slicing
+ * - Semantic similarity at different dimensions
+ */
+
+import { describe, it, expect, beforeAll } from 'vitest';
+import { createMRLEmbedder, createLegacyEmbedder, createEmbedder, VALID_DIMENSIONS } from '../lib/mrl-embedder.js';
+import { cosineSimilarity } from '../lib/utils.js';
+
+describe('MRL Embedder', () => {
+  let embedder256;
+  
+  beforeAll(async () => {
+    // Load embedder with 256d (default)
+    console.log('[Test] Loading MRL embedder (256d)...');
+    embedder256 = await createMRLEmbedder('nomic-ai/nomic-embed-text-v1.5', { dimension: 256 });
+    console.log('[Test] MRL embedder loaded');
+  }, 120000); // 2 min timeout for model download
+
+  describe('Dimension Configuration', () => {
+    it('should export valid dimensions', () => {
+      expect(VALID_DIMENSIONS).toEqual([64, 128, 256, 512, 768]);
+    });
+
+    it('should produce 256d embeddings by default', async () => {
+      const output = await embedder256('test text');
+      const vector = Array.from(output.data);
+      expect(vector.length).toBe(256);
+    });
+
+    it('should attach dimension metadata', () => {
+      expect(embedder256.dimension).toBe(256);
+      expect(embedder256.modelName).toBe('nomic-ai/nomic-embed-text-v1.5');
+    });
+  });
+
+  describe('Embedding Quality', () => {
+    it('should produce normalized vectors', async () => {
+      const output = await embedder256('normalized vector test');
+      const vector = Array.from(output.data);
+      const magnitude = Math.sqrt(vector.reduce((sum, v) => sum + v * v, 0));
+      expect(magnitude).toBeCloseTo(1, 3);
+    });
+
+    it('should generate different embeddings for different text', async () => {
+      const output1 = await embedder256('apple fruit');
+      const output2 = await embedder256('programming code');
+      
+      const vector1 = Array.from(output1.data);
+      const vector2 = Array.from(output2.data);
+      
+      const areSame = vector1.every((v, i) => Math.abs(v - vector2[i]) < 0.0001);
+      expect(areSame).toBe(false);
+    });
+  });
+
+  describe('Semantic Similarity', () => {
+    it('should give high similarity for semantically similar text', async () => {
+      const output1 = await embedder256('user authentication login');
+      const output2 = await embedder256('user login authentication');
+      
+      const vector1 = Array.from(output1.data);
+      const vector2 = Array.from(output2.data);
+      
+      const similarity = cosineSimilarity(vector1, vector2);
+      expect(similarity).toBeGreaterThan(0.85);
+    });
+    
+    it('should give lower similarity for different topics', async () => {
+      const output1 = await embedder256('database query SQL');
+      const output2 = await embedder256('pizza delivery food');
+      
+      const vector1 = Array.from(output1.data);
+      const vector2 = Array.from(output2.data);
+      
+      const similarity = cosineSimilarity(vector1, vector2);
+      expect(similarity).toBeLessThan(0.5);
+    });
+  });
+});
+
+describe('createEmbedder Factory', () => {
+  it('should create MRL embedder for nomic models', async () => {
+    const config = {
+      embeddingModel: 'nomic-ai/nomic-embed-text-v1.5',
+      embeddingDimension: 128,
+      device: 'cpu'
+    };
+    
+    const embedder = await createEmbedder(config);
+    expect(embedder.modelName).toContain('nomic');
+    expect(embedder.dimension).toBe(128);
+  }, 120000);
+
+  it('should fall back to legacy for MiniLM', async () => {
+    const config = {
+      embeddingModel: 'Xenova/all-MiniLM-L6-v2',
+      device: 'cpu'
+    };
+    
+    const embedder = await createEmbedder(config);
+    expect(embedder.dimension).toBe(384);
+  }, 120000);
+});