Latest updates to Meta’s code generation AI model, cordrama 70B, the “largest and best performing model” to date. The Code Llama tool was released in August and is free to use for both research and commercial use. According to Meta’s post on his AI blog, Code Llama 70B can handle more queries than previous versions, allowing developers to enter more prompts while programming and also improve accuracy.
Code Llama 70B recorded 53% accuracy on the HumanEval benchmark, better than GPT-3.5’s 48.1% and performing close to the 67% mark. OpenAI paper (PDF) Reported on GPT-4.
Built on Llama 2, Code Llama helps developers create code strings from prompts and debug human-written work. Last fall, Meta simultaneously released two of his Code Llama tools, Code Llama-Python and Code Llama-Instruct, which focus on specific coding languages.
Code Llama 70B is available in three versions of the code generator and remains free for research and commercial use. The large-scale model was trained on 1 TB of code and code-related data. It is hosted on the code repository Hugging Face and provides access to GPUs to run AI models.
Meta said the larger models, the 34B and 70B, “return the best results and allow for better coding assistance.”