Note: This library is still under active development and is not yet ready for production use.

An easy-to-sse Java SDK for running the LLaMA (v1 and v2) models and their variants on edge devices, powered by LLaMA.cpp.

Using llama4j with GGUF models is pretty easy. First add the related dependencies to your pom.xml:

<dependency>
    <groupId>org.javallm</groupId>
    <artifactId>llama4j</artifactId>
    <version>0.0.1</version>
</dependency>

Then, you should download GGML models from huggingface or somewhere else. Notice that only models with .gguf suffix since this is the latest format that the upstream LLaMA.cpp supports.

Now, you can happily play with your model! Here is a simple example:

SimpleCasualLM client = new SimpleCasualLM(GGML_MODEL_PATH);
client.infer("Once upon a time, there was a little girl named Lily.", System.out::print);