AiSee LLaMa Android SDK

An Android SDK for:

• Searching and retrieving .gguf LLaMA models from Hugging Face
• Running on-device inference with streamed responses

Designed for private, efficient, mobile-friendly AI deployments.

---

Installation via JitPack

1. Add JitPack to your root build.gradle:

allprojects {
    repositories {
        ...
        maven { url 'https://jitpack.io' }
    }
}

2. Add the library to your build.gradle (app module):

dependencies {
    implementation 'com.github.jaliyanimanthako:test-llama:v1.0.5'
}

---

🔍 Hugging Face Model Search (ai.aisee.llama.HF)

Programmatically search Hugging Face for .gguf models.

✅ Features

• Query Hugging Face with filters (author, tag, search)
• Automatically fetch and parse model metadata
• Support for pagination, sorting, and tree inspection (e.g., file structure)

HF Package Contents

File	Description
HFModelSearch.java	Main API for model search
HFModelInfo.java	Describes individual model metadata
HFModelTree.java	Fetches model file tree
HFModels.java	Aggregates API methods
HFEndpoints.java	Central endpoint constants
ExampleModel.java	Sample model structure
CustomDateDeserializer.java	Parses createdAt timestamps
CustomDateSerializer.java	Serializes timestamps for JSON

Usage Example

HFModelSearch modelSearch = new HFModelSearch();

List<HFModelSearch.ModelSearchResult> results = modelSearch.searchModels(
    "llama",                        // query
    "TheBloke",                     // author
    "gguf",                         // tag
    HFModelSearch.ModelSortParam.DOWNLOADS,
    HFModelSearch.ModelSearchDirection.DESCENDING,
    10,
    false,
    false
);

for (HFModelSearch.ModelSearchResult result : results) {
    Log.d("HF", result.modelId + " - " + result.description);
}

---

On-Device LLaMA Inference (ai.aisee.llama.LLaMa)

Run .gguf models with real-time streamed output.

✅ Features

• Load .gguf model from URI
• Dynamically set system prompts
• Receive partial responses with LiveData
• Final callback via LlamaListener
• Auto-formatting of <think> tags

LLaMa Package Contents

File	Description
ModelInference.java	Main interface for model interaction
ModelLoader.java	Handles loading .gguf model files
LLaMa.java	Core inference logic using native libraries
GGUFReader.java	Utilities for reading .gguf model metadata

Inference Usage

ModelInference model = ModelInference.getInstance(context);

model.setSystemPrompt("You're a helpful assistant.");
model.setListener(response -> Log.d("LLaMa", "Final: " + response));
model.partialResponse.observe(this, partial -> Log.d("LLaMa", "Partial: " + partial));
model.generateResponse("Where did I leave my book?");

Load a Model:

model.loadModel(modelUri,
    () -> Log.d("Model", "Model loaded successfully."),
    () -> Log.e("Model", "Failed to load model.")
);

---

Directory Structure

ai.aisee.llama
├── HF
│   ├── CustomDateDeserializer.java
│   ├── CustomDateSerializer.java
│   ├── ExampleModel.java
│   ├── HFEndpoints.java
│   ├── HFModelInfo.java
│   ├── HFModelSearch.java
│   ├── HFModelTree.java
│   └── HFModels.java
├── LLaMa
│   ├── GGUFReader.java
│   ├── LLaMa.java
│   ├── ModelInference.java
│   └── ModelLoader.java

---

Pro Tips

• Use Hugging Face search to dynamically discover models
• Store downloaded .gguf models locally and load via Uri
• Combine streamed LiveData responses with real-time UI feedback

---