Future Works

ZETIC.MLange is continuously evolving. We are actively researching the following technologies and will be updating them very soon.

1. LLM NPU Support

We are optimizing Large Language Models to fully utilize NPUs through advanced graph optimization and automated model splitting.

2. NPU Hardware Support Update

We are actively expanding our high-performance NPU support to a wider range of hardware platforms.

Target Platforms

• Mobile: Samsung Exynos, Google Edge TPU
• Microcontrollers: NXP, Infineon

3. On-device RAG (Sneak Peek)

Enhance model accuracy and relevance with local knowledge retrieval.

// Initialize RAG Pipeline
val rag = ZeticRAGPipeline(
    retriever = LocalVectorDB(path = "docs.db"),
    generator = ZeticMLangeModel("llama-3.2-1b")
)

// Retrieve and Generate
val response = rag.query("How do I use ZETIC.MLange?")

4. On-device MCP (Sneak Peek)

Standardized context management for robust AI agents.

// Define tool resource
val weatherTool = Tool("get_weather") { location -> 
    // ... implementation ...
}

// Pass tools to context
val agent = ZeticAgent(
    model = model,
    tools = listOf(weatherTool)
)

agent.chat("What's the weather in Seoul?")

5. On-device Multimodal NPU Accelerate Support

We are enabling support for Multimodal models that process text, images, and audio simultaneously through a unified encoder architecture. This allows for rich, interactive user experiences where the device can "see" and "hear" its environment.

Contact Us