Meta has introduced an updated AI model for code generation called Code Llama 70B, which is based on Llama 2. This is stated in the message of the company.

“We are releasing Code Llama 70B, the largest and best-performing model in the Code Llama family,” said Meta in a statement.

The updated AI model is available in three versions:

  • Code Llama – 70B, the foundational code model;
  • Code Llama – 70B – Python, 70B specialized for Python;
  • Code Llama – 70B – Instruct 70B, which is fine-tuned for understanding natural language instructions.

Code Llama 70B is trained on 1 trillion code tokens and associated data. Previous models were trained on 500 billion code and data tokens.

Meta

Meta also compared the accuracy of Code Llama 70B with other models. For example, in the HumanEval test, its accuracy reached 53%. For GPT 3.5, this figure was 48.1%, and for GPT 4 – 67%.