GitHub has announced the public preview of its new code completion model, GPT-4o Copilot, which is designed to enhance coding efficiency. The model is based on GPT-4o mini and has undergone mid-training on a code-focused corpus that exceeds 1 trillion tokens, utilizing reinforcement learning with code execution feedback (RLEF). It has been trained on over 275,000 public repositories across more than 30 programming languages, promising more accurate suggestions and improved performance for Visual Studio Code users. A live demo showcasing the capabilities of the new model is scheduled for February 19 at 12:30 PM EST, led by Doug Finke, a 16-time Microsoft MVP.
Perfect timing, had this GH Copilot demo planned for a week and now there is a new fine-tuned model! 🚨 Live Demo – Wed, Feb 19 @ 12:30 PM EST Join me, Doug Finke (16x Microsoft MVP), for AI-Assisted Coding: GitHub Copilot Essentials. See how AI transforms your dev workflow! https://t.co/jXLRstqqK1
GitHub Copilot 的代码自动完成(输入代码后灰色的提示部分)终于从 GPT-3.5 换到了微调后的 GPT-4o mini 模型,也就是 GPT-4o Copilot 模型,这个新模型基于 GPT-4o mini,额外训练了超过 275,000 个高质量的公共代码库,覆盖 30 多种主流编程语言。 不过要 VS Code… https://t.co/ZfEmdXsERV https://t.co/dPtoQqIvQH
VS Code users, meet GPT-4o Copilot! 👋 This new model is trained on 275,000+ public repos across 30+ languages, offering more accurate suggestions and improved performance. https://t.co/A2TGmoxMmy