Embeddings

// OpenAI embedding configuration
{
  modelName: "text-embedding-3-small",
  stripNewLines: true,
  batchSize: 512,
  dimensions: 1536
}

// From OpenAIEmbedding.ts
const obj: Partial<OpenAIEmbeddingsParams> = {
  openAIApiKey,
  modelName: "text-embedding-3-small"
}

if (stripNewLines) obj.stripNewLines = stripNewLines
if (batchSize) obj.batchSize = parseInt(batchSize, 10)
if (dimensions) obj.dimensions = parseInt(dimensions, 10)

const model = new OpenAIEmbeddings(obj)

// Azure OpenAI configuration
{
  modelName: "text-embedding-ada-002",
  azureOpenAIApiDeploymentName: "my-embedding-deployment",
  azureOpenAIApiVersion: "2024-02-15-preview",
  batchSize: 512
}

// Cohere configuration
{
  modelName: "embed-english-v3.0",
  inputType: "search_query" // or "search_document"
}

// Ollama configuration
{
  baseUrl: "http://localhost:11434",
  modelName: "nomic-embed-text"
}

// HuggingFace Inference configuration
{
  modelName: "sentence-transformers/all-MiniLM-L6-v2",
  apiKey: "hf_..."
}

// Google Generative AI configuration
{
  modelName: "embedding-001"
}

// Voyage AI configuration
{
  modelName: "voyage-2",
  inputType: "document" // or "query"
}

// Mistral AI configuration
{
  modelName: "mistral-embed"
}

// Together AI configuration
{
  modelName: "togethercomputer/m2-bert-80M-8k-retrieval"
}

// Reduce dimensions for storage efficiency
{
  modelName: "text-embedding-3-large",
  dimensions: 1024 // Original: 3072
}

{
  batchSize: 512, // Process 512 texts per API call
  stripNewLines: true // Remove newlines before embedding
}

// OpenAI-compatible endpoint
{
  basePath: "https://api.custom-endpoint.com/v1"
}

{
  timeout: 60000 // 60 seconds
}

// Optimize based on input size
{
  batchSize: 512 // For short texts (< 500 chars)
  batchSize: 256 // For medium texts (500-2000 chars)
  batchSize: 128 // For long texts (> 2000 chars)
}

// Use cache to avoid re-computing
const cache = new LocalFileCache("embeddings-cache")
const embeddings = new CachedEmbeddings({
  underlyingEmbeddings: new OpenAIEmbeddings(),
  cache
})

// Use reduced dimensions for large datasets
{
  modelName: "text-embedding-3-large",
  dimensions: 1024 // Reduced from 3072
}

// Embed a single query
const queryEmbedding = await embeddings.embedQuery(
  "What is machine learning?"
)

// Embed multiple documents
const docEmbeddings = await embeddings.embedDocuments([
  "Document 1 text",
  "Document 2 text",
  "Document 3 text"
])

// Upsert documents with embeddings
await vectorStore.addDocuments([
  { pageContent: "text", metadata: { source: "doc.pdf" } }
])

// Search with query
const results = await vectorStore.similaritySearch(
  "search query",
  4 // top K
)

// Implement custom embeddings
class CustomEmbeddings extends Embeddings {
  async embedDocuments(texts: string[]): Promise<number[][]> {
    // Your custom logic
    return embeddings
  }
  
  async embedQuery(text: string): Promise<number[]> {
    // Your custom logic
    return embedding
  }
}

Error: Vector dimension mismatch
Expected: 1536, Got: 768

// Check your embedding dimensions
OpenAI text-embedding-3-small: 1536
Ollama nomic-embed-text: 768

// Update vector store index accordingly

// Handle rate limits with retries
{
  timeout: 60000,
  maxRetries: 3,
  batchSize: 100 // Smaller batches
}

# Test OpenAI embeddings
curl https://api.openai.com/v1/embeddings \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input": "test", "model": "text-embedding-3-small"}'

# Test Ollama
curl http://localhost:11434/api/embeddings \
  -d '{"model": "nomic-embed-text", "prompt": "test"}'

// Reduce batch size for large documents
{
  batchSize: 50 // Smaller batches
}