Meta Launches Open-Source Code Llama to Challenge GitHub Copilot
Meta introduces Code Llama, an open-source, commercially viable large language model built specifically for programming tasks. The new AI family offers multiple model sizes and directly rivals existing tools like GitHub Copilot and OpenAI's Codex.
Meta releases Code Llama, a new open-source large language model designed specifically for programming and licensed for commercial use. This code-specialized version of LLaMA 2 supports software engineers by generating code, completing snippets, writing documentation, and debugging across popular languages like Python, C++, Java, and PHP.
The Code Llama family includes three main models with 7 billion, 13 billion, and 34 billion parameters, all trained on 500 billion tokens. The smaller models require fewer GPUs to run, with the 7-billion version operating on a single GPU, while every model in the family supports a massive context window of 100,000 tokens to process large sections of a codebase.
This release positions Meta as a direct competitor to Microsoft's GitHub Copilot and OpenAI's Codex in the coding assistant market. Additionally, Meta offers specialized fine-tuned versions for Python programming and natural language instruction, ensuring the AI provides safe and helpful responses when generating code from conversational prompts.