> For the complete documentation index, see [llms.txt](https://brindha.gitbook.io/mylearning/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://brindha.gitbook.io/mylearning/tools/codellama.md).

# CodeLlama

**What it is**

Code Llama is a code-specialized version of Llama 2, created by further training Llama 2 on code-specific datasets. It can generate code and natural language about code from both code and natural language prompts, and can also be used for code completion and debugging. 

**Model variants**

Code Llama is available in three distinct models: the foundational code model, a version explicitly fine-tuned for Python, and Code Llama Instruct designed to handle natural language instructions more effectively. It is available in 7B, 13B, and 34B parameter sizes. 

**Largest model**

Meta released Code Llama 70B, the largest model in the Code Llama family, in three versions: a foundational code model, a model specialized for Python, and a model fine-tuned for understanding natural language instructions. The 70B model works best for coding assistance, while smaller models are better suited for low-latency tasks like real-time code completion. 

**Performance**

Code Llama 70B handles more queries with an accuracy of 53% on the HumanEval benchmark, surpassing GPT-3.5's 48.1% and approaching the 67% mark achieved by GPT-4. 

**Supported languages**

Code Llama supports popular programming languages including Python, C++, Java, PHP, TypeScript, JavaScript, C#, and Bash, among others.

**Developer**

Meta AI, open-sourced and freely available for research and commercial use under the Code Llama Community License.