Llama Family FAQs

Question 1

What is the mission of Llama Family?

Accepted Answer

The mission of Llama Family is to promote the development of Artificial General Intelligence through open source. They aim to build an open platform for developers and enthusiasts to collaborate on creating an open-source ecosystem for Llama models.

Question 2

What types of models does Llama Family work with?

Accepted Answer

Llama Family works with various Llama models, including the original Llama models open-sourced by Meta (with sizes 7B, 13B, 34B, 70B), Code Llama models specialized for code, and Atom models which enhance Chinese language capabilities.

Question 3

What computing resources does Llama Family provide?

Accepted Answer

Llama Family provides access to GPU resources including GeForce RTX 30 and 40 Series, with NVIDIA H100 and A100 Tensor Core GPUs coming soon.

Question 4

What is Atom?

Accepted Answer

Atom is a model developed jointly by AtomEcho and Llama Family, based on the Llama architecture. It is trained on 2.7T of Chinese and multilingual corpora, with parameter sizes of 1B, 7B, and 13B, aimed at enhancing Chinese language capabilities.

Question 5

How many tokens was Llama 2 trained on?

Accepted Answer

Llama 2 was trained on 2.0T tokens across all its model sizes (7B, 13B, 34B, 70B).

Question 6

What are the different types of Code Llama models?

Accepted Answer

Code Llama comes in three types: Base Model (for general code generation tasks), Python Model (specialized for Python), and Instruct Model (fine-tuned with human instructions and self-instruct code synthesis data).

Llama Family Howto

More Information

How to Use Llama Family