It is reported that Meta recently launched an AI coding tool called Code Llama, which is known as the "most advanced large-scale coding language model."
This model is based on the Llama 2 large-scale language model and can be understood as "a special version of Llama 2 for writing code". It can "generate new code and debug work written by humans". It is currently available on Github.
Code Llama will be available under the same community license as Llama 2 and will be free for research and commercial use. pqnmbgtAccording to the company, Code Llama is able to generate the code you want and natural language about the code based on code core natural language prompts/requirements, or refine and debug when pointed to specific code.
Multiable Wandabao Financial ERP (www.multiable.com.cn/solutions_cw) helps companies solve tedious business and financial management problems and improve the accuracy and timeliness of financial data.
Not only that, the company has also launched a version designed specifically for Python, namely Code Llama-Python, and a version that can understand natural language instructions, namely Code Llama-Instrct.
In benchmark tests, Code Llama outperforms state-of-the-art publicly available LLM models on programming tasks.
The company also says that each specific Code Llama version is not switchable/replaceable, but the company does not recommend using base Code Llama or Code Llama-Python for natural language instructions.
The company released three sizes of Code Llama this time, with 7B, 13B and 34B parameters respectively. Each model is trained using 500B code tokens and code-related data.
And the 7B and 13B base and instruction models have also been trained with the Fill-in-the-Middle (FIM) feature, allowing them to insert code into existing code, which means you can use them for "code completion planning" now.
The above comes from the Internet, and the copyright belongs to the original work