Meta introduced the Llama 3.1 models in July 2024, which included the 8B, 70B and 405B versions. These multilingual large language models (LLMS) quickly became some of the most popular open source models available, amassing millions of downloads in a short time frame. On December 6 2024, Meta released the Llama 3.3 70B Instruct model. The instruction-tuned, generative text-only model has improvements in reasoning, math, general knowledge and instruction following, and it achieves similar performance as the Llama 3.1 405B model. The smaller model size makes fine-tuning and deployment of the model cost lower compared to its predecessor without sacrificing the performance and making it easier to adopt, because its smaller footprint can be supported by more GPU shapes.

The model is available to be downloaded from Meta’s website and on Hugging Face, an online model repository. Oracle Cloud Infrastructure (OCI) Data Science is a platform for data scientists and developers to work with open source models powered by OCI’s compute infrastructure with features that support the entire machine learning lifecycle. You can bring in Llama 3.3 70B Instruct from either Hugging Face or Meta to use inside OCI Data Science. OCI Data Science offers AI Quick Actions, a no-code solution for customers to seamlessly manage, deploy, fine-tune, and evaluate large language models inside our platform.

Working with Llama 3.3 model in OCI Data Science AI Quick Actions

You can access AI Quick Actions from a Data Science notebook. For more information on getting started with AI Quick Actions, see About AI Quick Actions. AI Quick Actions has a Hugging Face integration that allows users to easily bring in any model from Hugging Face, including Llama 3.3 70B. The process begins with registering the model in AI Quick Actions. Accessing the Llama 3.3 70B model in Hugging Face requires acceptance of Meta’s user agreement. Once you have been granted access to the model, you can generate a token from Hugging Face to log into the Hugging Face CLI in the terminal of the notebook you are using to access AI Quick Actions. Use the following command to validate your access: 

huggingface-cli login <your-hugging-face-token></your-hugging-face-token>

You can begin the model registration process by selecting the Register new model card in the Model Explorer. It opens the Model Registration panel. You can choose to register any model and search for the model by name. Alternatively, you can download the Llama 3.3 model artifact to your Object Storage. AI Quick Actions also supports bringing in a model from Object Storage. Go through the model registration process by choosing the I have artifacts in Object Storage option under Model Artifact on the Model Registration panel. After the model registration is complete, you can deploy and fine-tune the model with a service-managed container maintained by the Data Science service.

Get started with Llama 3.3 model in OCI Data Science

OCI Data Science enables customers to work with the latest Llama model and keep up with the rapidly evolving Generative AI space. To learn more about Oracle Cloud Infrastructure Data Science, visit our documentation and GitHub repository. To learn more about AI Quick Actions, see the documentation.