Ad

CodeGeeX: Multilingual Code Generation Model

CodeGeeX generates code in multiple languages. It performs crosslingual translation and acts as a customizable programming assistant. Supports 100+ languages.
Screenshot of zai-org/CodeGeeX homepage

CodeGeeX is a large-scale multilingual code generation model with 13 billion parameters, trained on a diverse code corpus encompassing over 20 programming languages. The model excels in generating executable programs and translating code between various languages. Its initial implementation focuses on improving performance and broader language support. It leverages a transformer architecture and is compatible with both Ascend and NVIDIA platforms.

CodeGeeX distinguishes itself through its multilingual capabilities, offering code generation and cross-lingual translation in numerous languages. The project is notable for being open-source and accessible via a free VS Code and Jetbrains extension. Performance benchmarks demonstrate superior accuracy compared to other open-source multilingual models. It also includes a benchmark dataset called HumanEval-X, furthering research efforts in multilingual code evaluation.

  • Multilingual Code Generation: Generates code in Python, C++, Java, JavaScript, Go, and 100+ other programming languages.
  • Crosslingual Code Translation: Translates code snippets between different languages with high accuracy.
  • VS Code/Jetbrains Extension: Provides a convenient extension for code completion, explanation, and summarization within popular IDEs.
  • Flexible Deployment: Supports both Ascend and NVIDIA platforms, allowing for deployment on various hardware setups.
  • Open-Source: All code and model weights are publicly available for research and development.

CodeGeeX is an active project with regular updates, including improved performance and new features. The project has a growing community support through Discord and other channels. Recent releases include model quantization, multi-GPU support, and wider IDE compatibility.Β  The inclusion of the paper accepted by KDD 2023 signifies research validation. The extension models are updated regularly, containing new features and device support.

Developers and researchers seeking a powerful multilingual code generation tool will find CodeGeeX valuable. It addresses the challenges of code understanding and translation across various programming languages, offering a streamlined coding experience. Unlike traditional single-language models, CodeGeeX facilitates rapid prototyping and cross-platform development by enabling developers to easily convert code between languages.

Summarize:
Share:
Stars
8,794
Forks
688
Issues
188
Created
3 years ago
Commit
1 year ago
License
APACHE-2.0
Archived
No
Updated 6 days ago

Similar Repositories