Awesome-AI-Coding

A list of AI coding topics.

Open a pull request to add or edit this list.

Projects

BigCode: open scientific collaboration run by Hugging Face.
Fauxpilot: Code completion server with CodeGen.
CodeGPT.nvim: ChatGPT in neovim.
org-ai: Emacs org-mode with OpenAI APIs.
Autodoc: Generate codebase documentation use LLM (OpenAI / Alpaca)
CodeAlpaca: LLaMA trained on code instruction following.
🐾 Tabby: An opensource / on-prem alternative to GitHub Copilot.
promptr: CLI tool to operating on your codebase using GPT.
ChatIDE: Extension let you talk to ChatGPT inside VSCode.
PromptMate: VSCode extension embed ChatGPT.
TurboPilot: CPU based copilot clone
CodeCapybara: Open Source LLaMA Model that Follow Instruction-Tuning for Code Generation.
CodeTF: A One-stop Transformer Library for State-of-the-art Code LLM
Rift: A opensource LSP leveraging edge language model.
Octopack
- OctoPack: Instruction Tuning Code Large Language Models
- Instruct fine-tuning Code LLMs on large scale github commit dataset.
Bloop: bloop is a (AI-powered) fast code search engine written in Rust.
Twinny: ollama based AI code completion plugin

Competition-level code generation with AlphaCode
RepoCoder: Repository-Level Code Completion Through Iterative Retrieval and Generation
- Combined LLM completion and CodeSearch
- CodeGen-350M + BoW based snippet search beat Codex
Repository-Level Prompt Generation for Large Language Models of Code
- Generate proposals candidates based with prios, e.g imports, files from same dirs.
- Use a proposal candidate classifier to select based proposals for LLM.
ML-Enhanced Code Completion Improves Developer Productivity
- 500M Encoder-Decoder based model, fine tuned on Google's monorepo.
- 34% acceptance rate for multi-line code completion suggestions.
Sparks of Artificial General Intelligence: Early experiments with GPT-4: Chapter 3 on coding scenario. Chat UX.
Efficient Training of Language Models to Fill in the Middle: Train decoder-only model with suffix context using a special token.
Toolformer: Language Models Can Teach Themselves to Use Tools: LLM as API glue layer.
CodeCompose: A Large-Scale Industrial Deployment of AI-assisted Code Authoring
- deployed as single line code completion to reduce latency to 300ms - 500ms.
- 1.3B parameter size.
- fine-tuning improves accuracy / bleu by 50% - 100%.