dev.langchain4j.model.embedding.onnx.allminilml6v2q.AllMiniLmL6V2QuantizedEmbeddingModel Maven / Gradle / Ivy

Go to download

Show more of this group Show more artifacts with this name
Show all versions of langchain4j-embeddings-all-minilm-l6-v2-q Show documentation

In-process all-minilm-l6-v2 (quantized) embedding model

There is a newer version: 1.0.0-alpha1

Show newest version

package dev.langchain4j.model.embedding.onnx.allminilml6v2q;

import dev.langchain4j.model.embedding.onnx.AbstractInProcessEmbeddingModel;
import dev.langchain4j.model.embedding.onnx.OnnxBertBiEncoder;
import dev.langchain4j.model.embedding.onnx.PoolingMode;

import java.util.concurrent.Executor;

import static dev.langchain4j.internal.ValidationUtils.ensureNotNull;

/**
 * Quantized SentenceTransformers all-MiniLM-L6-v2 embedding model that runs within your Java application's process.
 * 
 * Maximum length of text (in tokens) that can be embedded at once: unlimited.
 * However, while you can embed very long texts, the quality of the embedding degrades as the text lengthens.
 * It is recommended to embed segments of no more than 256 tokens.
 * 

 * Embedding dimensions: 384
 * 

 * Uses an {@link Executor} to parallelize the embedding process.
 * By default, uses a cached thread pool with the number of threads equal to the number of available processors.
 * Threads are cached for 1 second.
 * 
 * More details
 * here and
 * here
 */
public class AllMiniLmL6V2QuantizedEmbeddingModel extends AbstractInProcessEmbeddingModel {

    private static final OnnxBertBiEncoder MODEL = loadFromJar(
            "all-minilm-l6-v2-q.onnx",
            "all-minilm-l6-v2-q-tokenizer.json",
            PoolingMode.MEAN
    );

    /**
     * Creates an instance of an {@code AllMiniLmL6V2QuantizedEmbeddingModel}.
     * Uses a cached thread pool with the number of threads equal to the number of available processors.
     * Threads are cached for 1 second.
     */
    public AllMiniLmL6V2QuantizedEmbeddingModel() {
        super(null);
    }

    /**
     * Creates an instance of an {@code AllMiniLmL6V2QuantizedEmbeddingModel}.
     *
     * @param executor The executor to use to parallelize the embedding process.
     */
    public AllMiniLmL6V2QuantizedEmbeddingModel(Executor executor) {
        super(ensureNotNull(executor, "executor"));
    }

    @Override
    protected OnnxBertBiEncoder model() {
        return MODEL;
    }

    @Override
    protected Integer knownDimension() {
        return 384;
    }
}