GitHub Copilot AI creates open source alternatives – Visual Studio Magazine

0

New

GitHub Copilot AI creates open source alternatives

GitHub Copilot, described as an “AI pair programmer,” debuted this year with a splash, amazing developers with its ability to deliver chunks of code when a user types into Visual Studio Code and even generate entire applications only via typed commands.

Turn words into code
[Click on image for larger view.] Turn words into code (source: OpenAI).

That debut came in June, when Microsoft partner OpenAI announced the tool, powered by a new AI system called Codex, which has been described as an improved descendant of GPT-3 (Generative Pre-trained Transformer). able to translate natural language into code. Since then, it has been steadily improved and offered as an API.

GitHub co-pilot
[Click on image for larger view.] GitHub co-pilot (source: GitHub).

GitHub said that once the tech preview is finished testing and testing, the company intends to offer it as a commercial product for VS Code and the full-fledged Visual Studio IDE.

A paid product plan was apparently not suitable for some in the industry, as open source alternatives have sprung up.

Take, for example, GPT Code Clippy: the open source version of GitHub Copilot.

Demo of the VS Code Extension in animated action using one of the GPT Code Clippy Templates
[Click on image for larger view.] Demonstrate the VS code extension in an animated action using one of the GPT code Clippy models (source: Code.AI).

This wiki for the GPT-Code-Clippy (GPT-CC) project hosted in a GitHub repository says, “GPT-Code-Clippy (GPT-CC) is a community effort to create an open source version of GitHub Copilot, a Programmer of AI pairs based on GPT-3, called GPT-Codex. GPT-CC is refined on our GPT Code Clippy dataset from publicly available code on GitHub. It was created to make it easy for researchers to study large, deep learning models that are trained on code to better understand their capabilities and limitations. GPT-CC uses the GPT-Neo model as the base language model, which has been pre-trained on the Pile dataset and we use the objective of causal language modeling to train the model. ”

The “we” references above refer to the root of the repository, CodedotAI. One YouTube channel describes Code.AI as “a community dedicated to all things AI for code. In this community, we don’t just discuss deep learning or code generation, we also discuss things like scalable computing and code documentation! It’s a great place to find other like-minded researchers and developers, build a team of collaborators, find a project to work on, or brainstorm projects and research ideas! On this channel, we post video recordings of community events such as paper book clubs and podcasts! If you are interested in Code AI, please join us using this link : https://discord.gg/68NZFfxHxD! ”

The project’s GitHub repository explains the dataset criteria and search tool used for training, as well as the different GPT-CC models available, training details and more.

“Our ultimate goal is not only to develop an open source version of Github’s Code Copilot, but a version comparable in performance and ease of use,” says the wiki. “To this end, we are continually expanding our dataset and developing better models.” Along these lines, action items for the team include:

  • Pre-train the model from scratch with the dataset we organized from GitHub: we think this would be a pretty straightforward process if we have the IT resources.
  • Experiment with using GPT-J in code generation as recommended by evaluating large language models trained on the code
  • Extend the capabilities of GPT Code Clippy to other languages, especially under-represented ones
  • Design a custom loss function that penalizes non-compiling code
  • Design ways to update version and updates of programming languages

While GPT Code Clippy appears to be quite popular – nine backers, 207 stars, 20 forks – it’s not the only GitHub Copilot alternative to have seen the light of day since June.

The GPT-3 DEMO site, for example, lists GitHub Copilot and GPT-Code-Clippy, along with:

  • CodeVox: A speech and natural language code creation tool from Andrew Mayne, who works for OpenAI and created the project in a hackathon.
  • Tabnine: “Tabnine’s IDE AI code completion plug-in complements code based on millions of programs in all languages ​​and your own context, allowing developers to code better and faster. Deep Tabnine is based on GPT-2, which uses the Transformer network architecture. The architecture was first developed to solve natural language processing problems. While code modeling and natural language modeling may seem like independent tasks, code modeling requires understanding English in unexpected ways. “

Other media have published similar summaries of the Copilot alternatives, which appear to be mostly existing products. Drawn from several sources, these include:

  • Second Mate: “An open source mini-imitation of GitHub Copilot using EleutherAI GPT-Neo-2.7B (via Huggingface Model Hub) for Emacs. This is a much smaller model, so it probably won’t be as efficient. than Copilot, but can still be interesting to play with! “
  • Atom: Wikipedia: “Atom is a free, open source text and source code editor for macOS, Linux and Microsoft Windows with support for plugins written in JavaScript and built-in Git Control. Developed by GitHub, Atom is a desktop application built using web technologies. ”
  • Captain Stack: “This feature is somewhat similar to Github Copilot’s code suggestion. But instead of using AI, it sends your search query to Google, then fetches the StackOverflow responses and automatically completes them for you. “
  • YouCompleteMe: “a code completion engine for Vim.”
  • Clara: Analytics India Magazine says: “Clara is an alternative to Github Copilot for VSCode. As for the features, it supports almost 50 programming languages ​​and gives developers the snippers in an instant. Check out the source code on Github.
  • Kite: “Kite adds AI-powered code completions to your code editor, giving developers superpowers.”
  • Asm-Dude: “Assembly Syntax Highlight and Code Assistance for Assembly Source Files and Disassembly Window for Visual Studio 2015, 2017, and 2019. This extension can be found in the Extensions Gallery of Visual Studio or download the latest AsmDude.vsix installer (v1.9.6 .14). If assembly is too complicated but you still want to access machine-specific instructions, consider Intrinsics-Dude. “This comes as an extension of visual studio.

Other products that have been identified as alternatives to GitHub Copilot include Make, Spacemacs, Rust-analyzer and more, with some experts and sites bundling long-standing tools with open source counterfeits of AI-driven GitHub Copilot. , thus providing many functionalities. called options which at first glance are far below the capabilities of GitHub Copilot. So take them with a grain of salt.

Meanwhile, a lot of buzz still surrounds GitHub Copilot, which has been decried by the nonprofit FSF (Free Software Foundation) as “unacceptable and unfair” and which has caused existential angst among developers who fear their work will fail. be replaced by advanced AI coding systems, along the lines of: “Build an ASP.NET Core MVC website optimized for selling cars.”

Security and ethics concerns have also been raised about GitHub Copilot, so it will be worth checking out the final product when it comes out of the Tech Preview.

About the Author

David Ramel is editor and writer for Converge360.


Source link

Leave A Reply

Your email address will not be published.