This repository provides a collection of examples, tutorials, and applications for developers leveraging Liquid AI's open-weight Foundational Models (LFM) and the open-source LEAP SDK. It targets developers aiming to build applications with on-device AI capabilities, offering resources for model fine-tuning, edge deployment, and end-to-end solution development, thereby enabling efficient local AI workflows.

How It Works

The project centers around the LFM2 series of text-to-text models (350M to 8B parameters) and LFM2-VL vision-language models, both designed for on-device deployment. These models are optimized for agentic tasks, data extraction, Retrieval-Augmented Generation (RAG), and multi-turn conversations. The LEAP Edge SDK, a native framework for Android (Kotlin) and iOS (Swift), facilitates seamless integration and deployment of these LFM2 models onto mobile devices, abstracting the complexity of local inference.

Quick Start & Requirements

While specific installation commands and version requirements (e.g., Python, CUDA) are not detailed, the project focuses on using open-weight models and an open-source SDK. Key resources include official documentation at https://leap.liquid.ai/docs, community support via Discord, and access to tutorials and example code repositories. The emphasis on "on-device" deployment suggests local execution capabilities without necessarily requiring high-end server hardware, though specific resource footprints are not provided.

Highlighted Details

• LFM2 Models: Offer a range of sizes (350M-8B parameters) suitable for agentic tasks, data extraction, RAG, and conversations, with specialized Nano checkpoints (e.g., LFM2-1.2B-Extract, LFM2-1.2B-RAG, LFM2-1.2B-Tool, LFM2-350M-Math) pre-trained for specific use cases.
• Vision-Language Capabilities: The LFM2-VL series (e.g., LFM2-VL-1.6B, LFM2-VL-450M) enables text and image input for text generation, also optimized for edge devices.
• LEAP Edge SDK: Simplifies on-device LLM deployment for mobile developers, offering an API experience comparable to cloud-based LLMs.
• Multi-language Support: Models natively support English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese, and Spanish, with community efforts underway to expand language coverage.

Maintenance & Community

The project actively fosters community engagement through a Discord server (Join our community) and encourages contributions by submitting pull requests with links to community-built project repositories. Comprehensive documentation is available at https://leap.liquid.ai/docs.

Licensing & Compatibility

The provided README content does not specify the licensing terms for the LFM models, the LEAP SDK, or the repository itself. This lack of explicit licensing information presents a potential adoption blocker, particularly for commercial use or integration into closed-source projects.

Limitations & Caveats

The LFM2 models are explicitly not recommended for knowledge-intensive tasks or those requiring programming skills. The README does not detail specific hardware prerequisites (e.g., RAM, GPU) for running models locally or on edge devices, nor does it provide performance benchmarks. The absence of clear licensing information is a significant caveat.