nllg/TikZilla-8B-RL

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Mar 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

TikZilla-8B-RL is an 8 billion parameter language model developed by nllg, specifically designed for generating TikZ/LaTeX figures from natural language descriptions. This model is based on TikZilla-8B and has been further optimized using reinforcement learning (RL) on the DaTikZ-V4 dataset. It excels at translating textual figure requirements into precise, ready-to-use LaTeX code, making it ideal for scientific documentation and academic publishing.

Loading preview...

TikZilla-8B-RL: LaTeX Figure Generation

TikZilla-8B-RL is an 8 billion parameter language model developed by nllg, specialized in converting natural language descriptions into TikZ/LaTeX figures. This model builds upon the TikZilla-8B architecture and significantly enhances its capabilities through reinforcement learning (RL) applied to the DaTikZ-V4 dataset.

Key Capabilities

  • TikZ/LaTeX Code Generation: Translates detailed natural language prompts into complete and valid LaTeX documents containing TikZ figures.
  • Scientific Figure Specialization: Optimized for creating various scientific plots and diagrams, including line plots with specific labels, ranges, and styles.
  • Reinforcement Learning Enhanced: Benefits from RL training on a dedicated dataset (DaTikZ-V4) to improve the accuracy and quality of generated figures.

Good For

  • Automating Figure Creation: Ideal for researchers, academics, and technical writers who need to quickly generate complex scientific figures without manual TikZ coding.
  • Streamlining Documentation: Integrates seamlessly into LaTeX-based workflows for papers, theses, and reports.
  • Prototyping Visualizations: Enables rapid prototyping of visual concepts by describing them in plain English and receiving executable LaTeX code.