CtrlK
BlogDocsLog inGet started
Tessl Logo

tessl/maven-dev-langchain4j--langchain4j-vertex-ai

LangChain4j integration for Google Vertex AI models including chat, language, embedding, image, and scoring capabilities

—

Pending

Quality

Pending

Does it follow best practices?

Impact

Pending

No eval scenarios have been run

Overview
Eval results
Files

examples.mddocs/models/embedding/

Embedding Model Examples

Basic Embedding

import dev.langchain4j.model.vertexai.VertexAiEmbeddingModel;
import dev.langchain4j.data.segment.TextSegment;
import dev.langchain4j.data.embedding.Embedding;
import dev.langchain4j.model.output.Response;

VertexAiEmbeddingModel model = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .build();

TextSegment segment = TextSegment.from("Machine learning is a subset of AI");
Response<Embedding> response = model.embed(segment);
Embedding embedding = response.content();
float[] vector = embedding.vector();
System.out.println("Dimension: " + embedding.dimension());

Embed Multiple Texts

List<TextSegment> segments = List.of(
    TextSegment.from("First document about AI"),
    TextSegment.from("Second document about machine learning"),
    TextSegment.from("Third document about neural networks")
);

Response<List<Embedding>> response = model.embedAll(segments);
List<Embedding> embeddings = response.content();

for (int i = 0; i < embeddings.size(); i++) {
    System.out.println("Document " + i + " embedding: " +
                       embeddings.get(i).dimension() + " dimensions");
}

With Task Type Optimization

// For document indexing
VertexAiEmbeddingModel documentModel = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .taskType(VertexAiEmbeddingModel.TaskType.RETRIEVAL_DOCUMENT)
    .build();

// For search queries
VertexAiEmbeddingModel queryModel = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .taskType(VertexAiEmbeddingModel.TaskType.RETRIEVAL_QUERY)
    .build();

Advanced Configuration

VertexAiEmbeddingModel model = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .taskType(TaskType.SEMANTIC_SIMILARITY)
    .maxSegmentsPerBatch(100)      // Process 100 segments per API call
    .maxTokensPerBatch(10000)       // Max 10k tokens per batch
    .outputDimensionality(256)      // Reduce to 256 dimensions
    .autoTruncate(true)             // Auto-truncate long texts
    .titleMetadataKey("doc_title")  // Custom title metadata key
    .maxRetries(3)
    .build();

Using Predefined Model Names

import dev.langchain4j.model.vertexai.VertexAiEmbeddingModelName;

VertexAiEmbeddingModel model = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName(VertexAiEmbeddingModelName.TEXT_EMBEDDING_004.toString())
    .build();

// Check dimension
Integer dimension = VertexAiEmbeddingModelName.TEXT_EMBEDDING_004.dimension();
System.out.println("Expected dimension: " + dimension);

Calculate Token Counts

List<TextSegment> segments = List.of(
    TextSegment.from("Short text"),
    TextSegment.from("A much longer text with many more tokens to process")
);

List<Integer> tokenCounts = model.calculateTokensCounts(segments);
for (int i = 0; i < tokenCounts.size(); i++) {
    System.out.println("Segment " + i + " has " + tokenCounts.get(i) + " tokens");
}

Using with Metadata

import dev.langchain4j.data.document.Metadata;

TextSegment segment = TextSegment.from(
    "Document content here",
    Metadata.from("title", "Important Document")
);

Response<Embedding> response = model.embed(segment);

Custom Title Metadata Key

VertexAiEmbeddingModel model = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .titleMetadataKey("doc_title")  // Custom metadata key
    .build();

TextSegment segment = TextSegment.from(
    "Document content",
    Metadata.from("doc_title", "My Document")  // Use same key
);

Response<Embedding> response = model.embed(segment);

Semantic Search Example

// Index documents
List<TextSegment> documents = List.of(
    TextSegment.from("Python is a programming language"),
    TextSegment.from("Java is used for enterprise applications"),
    TextSegment.from("Machine learning uses neural networks")
);

Response<List<Embedding>> docEmbeddings = documentModel.embedAll(documents);

// Search query
TextSegment query = TextSegment.from("programming languages");
Response<Embedding> queryEmbedding = queryModel.embed(query);

// Calculate similarities (cosine similarity)
for (int i = 0; i < docEmbeddings.content().size(); i++) {
    double similarity = cosineSimilarity(
        queryEmbedding.content().vector(),
        docEmbeddings.content().get(i).vector()
    );
    System.out.println("Document " + i + " similarity: " + similarity);
}

Reduced Dimensionality

VertexAiEmbeddingModel model = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .outputDimensionality(256)  // Reduce from 768 to 256
    .build();

Response<Embedding> response = model.embed(TextSegment.from("Some text"));
System.out.println("Dimension: " + response.content().dimension()); // 256

Auto-Truncate Long Texts

VertexAiEmbeddingModel model = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .autoTruncate(true)  // Automatically truncate if text too long
    .build();

String veryLongText = "...";  // Text longer than model limit
TextSegment segment = TextSegment.from(veryLongText);
Response<Embedding> response = model.embed(segment);  // No error, auto-truncated

Batch Processing with Limits

// Configure batching
VertexAiEmbeddingModel model = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .maxSegmentsPerBatch(50)   // Process 50 at a time
    .maxTokensPerBatch(5000)    // Max 5000 tokens per batch
    .build();

// Large list of documents
List<TextSegment> manyDocuments = // ... 1000 documents
Response<List<Embedding>> response = model.embedAll(manyDocuments);
// Automatically batched into multiple API calls

Multilingual Embedding

VertexAiEmbeddingModel multilingualModel = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-multilingual-embedding-002")
    .build();

List<TextSegment> multilingualTexts = List.of(
    TextSegment.from("Hello world"),           // English
    TextSegment.from("Bonjour le monde"),      // French
    TextSegment.from("Hola mundo")             // Spanish
);

Response<List<Embedding>> response = multilingualModel.embedAll(multilingualTexts);

Error Handling

try {
    Response<List<Embedding>> response = model.embedAll(segments);
    List<Embedding> embeddings = response.content();
} catch (Exception e) {
    System.err.println("Embedding failed: " + e.getMessage());
    // Automatic retries already attempted up to maxRetries
}

RAG Implementation Example

// 1. Create embedding model for documents
VertexAiEmbeddingModel docModel = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .taskType(TaskType.RETRIEVAL_DOCUMENT)
    .build();

// 2. Index knowledge base
List<TextSegment> knowledgeBase = loadDocuments();
Response<List<Embedding>> docEmbeddings = docModel.embedAll(knowledgeBase);

// 3. Create embedding model for queries
VertexAiEmbeddingModel queryModel = VertexAiEmbeddingModel.builder()
    .endpoint("us-central1-aiplatform.googleapis.com:443")
    .project("your-project-id")
    .location("us-central1")
    .publisher("google")
    .modelName("text-embedding-004")
    .taskType(TaskType.RETRIEVAL_QUERY)
    .build();

// 4. Search
String userQuery = "How does photosynthesis work?";
Response<Embedding> queryEmbedding = queryModel.embed(TextSegment.from(userQuery));

// 5. Find most similar documents
List<TextSegment> relevantDocs = findTopK(
    docEmbeddings.content(),
    queryEmbedding.content(),
    knowledgeBase,
    3  // Top 3
);

Install with Tessl CLI

npx tessl i tessl/maven-dev-langchain4j--langchain4j-vertex-ai@1.11.0

docs

models

embedding

api.md

examples.md

overview.md

index.md

quick-reference.md

tile.json