🚀 CodeGenie – Your AI-Powered Coding Assistant
📌 Introduction
Welcome to CodeGenie, an intelligent VS Code extension that brings the power of AI-driven code generation right into your workflow! Powered by DeepSeek Coder and accelerated by an RTX 4090 GPU, this extension provides real-time code suggestions, autocompletions, and context-aware snippets, making development faster, smarter, and more efficient.
🎯 Purpose of the Project
- 🧠 AI-Powered Code Assistance – Integrate a cutting-edge large language model (LLM) to generate intelligent and contextually relevant code snippets.
- 🔌 Seamless VS Code Integration – Develop a fully functional VS Code extension that enhances coding productivity.
- ⚡ Optimized Performance with GPU Acceleration – Utilize a high-performance RTX 4090 GPU to ensure fast and efficient AI inference.
- 🚀 Enhancing Developer Experience – Automate repetitive tasks, reduce errors, and speed up the coding process with AI-driven suggestions.
🌍 Applications of the Project
✅ Real-Time Code Autocompletion – Get instant code suggestions while typing, reducing keystrokes and enhancing efficiency.
✅ AI-Generated Code Snippets – Automatically generate boilerplate code and reusable functions based on contextual prompts.
✅ Multi-Language Support – Assist developers across various programming languages, improving cross-language compatibility.
✅ Error Detection & Fix Suggestions – Help identify common coding mistakes and offer AI-powered solutions.
✅ Increased Productivity – Reduce development time, improve code quality, and streamline workflows with AI automation.
🚀 Architecture Diagram
🧠 WorkFlow
📖 Brief Explanation of the Reference Research Papers
🔬 DeepSeek-Coder: Open-Source Large Language Models for Code Intelligence
DeepSeek-Coder is a series of open-source large language models (LLMs) designed for code understanding, generation, and completion. These models, ranging from 1.3B to 33B parameters, are pre-trained on a high-quality, project-level code corpus and employ advanced learning strategies to achieve state-of-the-art performance in open-source code intelligence.
🏆 Key Contributions and Features
- Scalable Model Variants – Supports models in sizes of 1.3B, 6.7B, and 33B parameters, catering to various computational and application needs.
- Extensive Pretraining – Trained on 2 trillion tokens from 87 programming languages, ensuring robust cross-language code comprehension.
- Repository-Level Data Organization – Unlike conventional file-based training, DeepSeek-Coder incorporates repository-level data structuring, significantly enhancing cross-file code generation.
- Advanced Code Completion & Infilling – Implements a fill-in-the-middle (FIM) task with a 16K context window, improving in-line completions and cross-file dependency resolution.
- Superior Open-Source Performance – Outperforms other open-source models (e.g., CodeLlama, StarCoder, CodeGeeX2) and even surpasses GPT-3.5-Turbo in multiple code-related benchmarks.
- Instruction-Tuned Model – DeepSeek-Coder-Instruct further fine-tuned with high-quality instructional data, improving performance in multi-turn coding tasks and real-world programming challenges.
- Long-Context Adaptation – Utilizes an optimized RoPE-based positional encoding, enabling effective processing of up to 16K tokens, with the potential to scale beyond 64K tokens.
- Cross-File Code Completion – Evaluated on the CrossCodeEval benchmark, demonstrating superior performance in handling multi-file codebases.
- Mathematical & Reasoning Capabilities – Fine-tuned versions (DeepSeek-Coder-v1.5) incorporate additional pretraining for mathematical reasoning and natural language understanding tasks.
- HumanEval & MBPP – Achieves 66.0% accuracy, outperforming CodeLlama-33B.
- LeetCode Contest Benchmark – Surpasses GPT-3.5-Turbo and is the first open-source model to outperform it.
- DS-1000 (Data Science Benchmark) – Excels in applying libraries like NumPy, Pandas, and PyTorch to real-world data science problems.
- CrossCodeEval – Demonstrates state-of-the-art performance in cross-file code understanding, a key factor for repository-level code comprehension.
📜 License & Availability
DeepSeek-Coder is released under a permissive open-source license, allowing unrestricted use for research and commercial applications.
🔗 GitHub Repository: DeepSeek-Coder
Requirements and Instructions
- requirements.txt -Initially install all the requirements from requirements.txt using the command pip install requirements.txt
- download_model.sh -Refer to the scripts/download_model.sh file to ensure the download of Deepseek Instruct 1.3b instruct model.
👥 Contributors Overview
👥 Contributors Overview
🔥 Get ready to build the future of AI-powered development with CodeGenie! 🚀
