Meta has developed Code Llama, a state-of-the-art language model, aimed at generating code and assisting with coding tasks, and it is now available through Amazon’s SageMaker JumpStart. Code Llama operates in Python, C++, Java, PHP, C#, TypeScript, and Bash, with the aim of boosting developers’ productivity and streamlining software processes. The model explores three variations and four sizes to cater to a broad array of applications. These were designed using Llama 2 and trained on 500 billion tokens of code data.
Amazon’s SageMaker JumpStart allows developers access to many models including Code Llama. These models, often called foundation models, are large-scale and can be applied in a variety of tasks like text summarization, digital art creation, and language translation. Because these models can be costly to train, using pre-existing pre-trained foundation models saves users’ time and resources.
To use the Code Llama model on SageMaker, developers go through a simple process via SageMaker Studio, which includes searching for the Code Llama models, choosing the desired model, agreeing with the End User License Agreement (EULA) and clicking on “Deploy”. Users can also deploy the model using a SageMaker Python SDK, after which they can execute inferences using Boto3 or the SageMaker Python SDK.
Amazon’s SageMaker can act as a hub for machine learning, giving developers access to models with virtually one click. The new Code Llama models promise more efficient programming tasks, and developers are encouraged to use them with their own datasets and coding use cases.
The news was announced by Dr. Kyle Ulrich, an Applied Scientist with the Amazon SageMaker JumpStart team, Dr. Farooq Sabir, a Senior Artificial Intelligence and Machine Learning Specialist Solutions Architect at AWS, and June Won, a product manager with SageMaker JumpStart.
