TikZilla-3B Overview

TikZilla-3B is a specialized 3.1 billion parameter language model, built upon the Qwen2.5-3B architecture, designed for the unique task of generating TikZ/LaTeX figures from natural language input. Developed by nllg, this model was fine-tuned using supervised fine-tuning (SFT) on the DaTikZ-V4 dataset, specifically targeting the creation of scientific and technical diagrams.

Key Capabilities

• Natural Language to TikZ/LaTeX Conversion: Translates detailed text descriptions into executable LaTeX code for vector graphics.
• Scientific Figure Generation: Optimized for creating common scientific plots, diagrams, and illustrations.
• High Context Length: Supports a context window of 32768 tokens, allowing for complex and detailed figure descriptions.
• Complete LaTeX Output: Generates full LaTeX documents containing the TikZ figure, ready for compilation.

Good for

• Automating Figure Creation: Researchers and technical writers can quickly generate figures without manual TikZ coding.
• Academic Publishing: Ideal for producing high-quality, customizable vector graphics for papers, theses, and presentations.
• Prototyping Diagrams: Rapidly create visual representations from textual ideas.
• Educational Tools: Assisting in the creation of illustrative materials for STEM education.