TIGER-Lab/VisCoder2-14B
VisCoder2-14B by TIGER-Lab is a 14B parameter multi-language visualization coding model based on Qwen2.5-Coder-14B-Instruct. It is specifically trained for executable code generation, rendering, and iterative self-debugging across 12 programming languages. This model excels at generating semantically consistent visual outputs by aligning natural-language instructions with rendering results, making it ideal for complex visualization tasks.
Loading preview...
VisCoder2-14B: Multi-Language Visualization Coding Agent
VisCoder2-14B, developed by TIGER-Lab, is a specialized 14-billion parameter model built upon Qwen2.5-Coder-14B-Instruct. Its core innovation lies in its ability to generate executable visualization code across 12 programming languages, focusing on both successful execution and semantic consistency of visual outputs.
Key Capabilities
- Multi-Language Visualization Code Generation: Generates code for visualization tasks in 12 different programming languages.
- Executable Code & Rendering: Focuses on producing code that not only runs but also renders visually accurate results.
- Iterative Self-Debugging: Designed to iteratively debug its own generated code, improving output quality.
- Trained on VisCode-Multi-679K: Leverages a large-scale instruction-tuning dataset specifically for executable visualization tasks.
- Strong Performance on VisPlotBench: Achieves notable improvements, especially in multi-round self-debugging scenarios, across 8 languages on the VisPlotBench evaluation suite.
Good For
- Developers and researchers needing to generate visualization code from natural language descriptions.
- Applications requiring robust, executable, and visually accurate code outputs in multiple programming environments.
- Tasks benefiting from an agent that can self-correct and debug visualization code.