ali-thowfeek/Charty-1B
Charty-1B by ali-thowfeek is a 1.17 billion parameter text-to-Mermaid diagram generation model, fine-tuned from Liquid AI's LFM2.5-1.2B-Instruct. Utilizing a hybrid LFM2 architecture, it specializes in converting natural language prompts into raw Mermaid syntax for various diagram types. This compact model is optimized for on-device and mobile deployment, enabling offline diagram-as-code generation on low-end hardware.
Loading preview...
Charty-1B: Compact Text-to-Mermaid Diagram Generation
Charty-1B is a specialized 1.17 billion parameter language model developed by ali-thowfeek, designed for efficient text-to-Mermaid diagram generation. Fine-tuned from Liquid AI's LFM2.5-1.2B-Instruct, it leverages a hybrid LFM2 architecture combining 10 double-gated LIV convolution blocks and 6 GQA attention blocks.
Key Capabilities
- Direct Mermaid Syntax Output: Generates raw Mermaid syntax (flowcharts, sequence, class, state, ER, Gantt, pie, git graphs, mindmaps, quadrant charts) directly from natural language prompts, without conversational filler or markdown fences.
- Compact & Efficient: With 1.17B parameters and a 32,768-token context length, it's optimized for deployment on low-end hardware, including mobile phones, edge devices, laptops, and embedded systems.
- Multilingual Support: Inherits multilingual capabilities from its base model, supporting English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.
- Quantized Versions Available: Offers F16, Q8_0, and Q4KM quantized GGUF versions for CPU-only inference and further memory optimization, ideal for offline and private diagram generation applications.
Intended Use Cases
- On-device Diagram Generation: Powering applications that require offline, private, and fast diagram-as-code generation directly on user devices.
- Edge Computing: Deploying diagram generation capabilities on edge devices with limited computational resources.
- Developer Tools: Integrating into IDEs or documentation tools for quick diagram creation from text descriptions.
Charty-1B was fine-tuned using LoRA (SFT) via Unsloth and Hugging Face TRL on the ali-thowfeek/text-to-mermaid dataset, ensuring 100% of training examples produce valid Mermaid v11 core syntax.