Loke-60000/rin-mobile-preview
Loke-60000/rin-mobile-preview is a 5.1 billion parameter multimodal language model designed for on-device agentic workloads. Trained on approximately 895,000 tokens, it features a consistent voice named Rin and supports text-to-text, image-to-text, and speech-to-text capabilities. This compact model is optimized for running directly on mobile devices or laptops without cloud dependency, excelling in long-horizon agentic tasks, reasoning, and tool calls.
Loading preview...
Overview
Loke-60000/rin-mobile-preview is a 5.1 billion parameter multimodal model specifically engineered for efficient on-device execution. It is designed to perform agentic work directly on hardware like phones or laptops, eliminating the need for server or cloud infrastructure. The model was trained on approximately 895,000 tokens to cultivate a consistent and clear voice, referred to as Rin.
Key Capabilities
- Multimodal Input: Processes and understands text, images, and audio.
- Text-to-Text: Handles chat, coding assistance, technical support, and complex, multi-step agentic tasks.
- Image-to-Text: Describes and reasons about visual content from images.
- Speech-to-Text: Transcribes audio clips and can answer questions based on their content.
- Agentic Features: Supports private, step-by-step reasoning and tool calls, making it suitable for autonomous task execution.
On-Device Optimization
This model is quantized for phone-class hardware, with a size of approximately 4.4 GB, making it highly suitable for local deployment. Its design prioritizes running complex AI tasks without external dependencies, offering privacy and low-latency performance for mobile and edge computing applications.