Skip to main content

Vearch

https://github.com/vearch/vearch

Maven Dependency​

note

Since 0.37.0, langchain4j-vearch has migrated to langchain4j-community and is renamed to langchain4j-community-vearch.

0.36.2 and previous:

<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-vearch</artifactId>
    <version>0.36.2</version>
</dependency>

0.37.0 and later:

<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-community-vearch</artifactId>
    <version>0.37.0</version>
</dependency>

Or, you can use BOM to manage dependencies consistently:

<dependencyManagement>
    <dependency>
        <groupId>dev.langchain4j</groupId>
        <artifactId>langchain4j-community-bom</artifactId>
        <version>0.37.0</version>
        <typ>pom</typ>
        <scope>import</scope>
    </dependency>
</dependencyManagement>

APIs​

0.36.2 and previous​

note
  • 0.36.2 and previous langchain4j-vearch uses Vearch old api, which is deprecated in vearch 3.4.x version.
  • 0.37.0 and later langchain4j-community-vearch uses Vearch latest api, which supports 3.5.x and 3.4.x version.

We recommend to use langchain4j-community-vearch which supports the latest version of Vearch.

To use VearchEmbeddingStore, you need to instantiate a VearchConfig:

String embeddingFieldName = "text_embedding";
String textFieldName = "text";
Map<String, Object> metadata = createMetadata().toMap();

// init properties
Map<String, SpacePropertyParam> properties = new HashMap<>(4);
properties.put(embeddingFieldName, SpacePropertyParam.VectorParam.builder()
        .index(true)
        .storeType(SpaceStoreType.MEMORY_ONLY)
        .dimension(384)
        .build());
properties.put(textFieldName, SpacePropertyParam.StringParam.builder().build());
// put metadata... e.g. properties.put("name", SpacePropertyParam.StringParam.builder().build());

VearchConfig vearchConfig = VearchConfig.builder()
        .spaceEngine(SpaceEngine.builder()
                .name("gamma")
                .indexSize(1L)
                .retrievalType(RetrievalType.FLAT)
                .retrievalParam(RetrievalParam.FLAT.builder()
                        .build())
                .build())
        .properties(properties)
        .embeddingFieldName(embeddingFieldName)
        .textFieldName(textFieldName)
        .databaseName(databaseName)
        .spaceName(spaceName)
        .modelParams(singletonList(ModelParam.builder()
                .modelId("vgg16")
                .fields(singletonList("string"))
                .out("feature")
                .build()))
        .build();

Then, you can create a VearchEmbeddingStore:

VearchEmbeddingStore embeddingStore = VearchEmbeddingStore.builder()
        .vearchConfig(vearchConfig)
        .baseUrl(baseUrl)
        .build();

0.37.0 and later​

To use VearchEmbeddingStore, you need to instantiate a VearchConfig:

String embeddingFieldName = "text_embedding";
String textFieldName = "text";
String spaceName = "embedding_space_" + ThreadLocalRandom.current().nextInt(0, Integer.MAX_VALUE);

// init Fields
List<Field> fields = new ArrayList<>(4);
List<String> metadataFieldNames = new ArrayList<>();
fields.add(VectorField.builder()
        .name(embeddingFieldName)
        .dimension(embeddingModel.dimension())
        .index(Index.builder()
                .name("gamma")
                .type(IndexType.HNSW)
                .params(HNSWParam.builder()
                        .metricType(MetricType.INNER_PRODUCT)
                        .efConstruction(100)
                        .nLinks(32)
                        .efSearch(64)
                        .build())
                .build())
        .build()
);
fields.add(StringField.builder().name(textFieldName).fieldType(FieldType.STRING).build());
// put metadata... e.g. fields.add(StringField.builder().name("name").fieldType(FieldType.STRING).build());

VearchConfig vearchConfig = VearchConfig.builder()
        .databaseName(databaseName)
        .spaceName(spaceName)
        .textFieldName(textFieldName)
        .embeddingFieldName(embeddingFieldName)
        .fields(fields)
        .metadataFieldNames(metadataFieldNames)
        .searchIndexParam(HNSWSearchParam.builder()
                // Only support INNER_PRODUCT now
                .metricType(MetricType.INNER_PRODUCT)
                .efSearch(64)
                .build())
        .build();

Then, you can create a VearchEmbeddingStore:

VearchEmbeddingStore embeddingStore = VearchEmbeddingStore.builder()
        .vearchConfig(vearchConfig)
        .baseUrl(baseUrl)
        .logRequests(true)
        .logResponses(true)
        .build();

Examples​