Llama 2 model sizes (7B, 13B, 70B)

All three currently available Llama 2 model sizes (7B, 13B, 70B) are trained on 2 trillion tokens and have double the context length of Llama 1. Llama 2 encompasses a series of generative text models that have been pretrained and fine-tuned, varying in size from 7 billion to 70 billion parameters.

Meta’s specially fine-tuned models (Llama-2-Chat) are tailored for conversational scenarios. In most of our benchmark tests, Llama-2-Chat models surpass other open-source chatbots and match the performance and safety of renowned closed-source models such as ChatGPT and PaLM.

Differences between Llama 2 models (7B, 13B, 70B)

Llama 2 7b is swift but lacks depth, making it suitable for basic tasks like summaries or categorization.

Llama 2 13b strikes a balance: it’s more adept at grasping nuances compared to 7b, and while it’s less cautious about potentially offending, it’s still quite conservative. This variant excels in creative endeavors such as crafting stories or poetry, even if slightly slower than 7b.

Llama 2 70b stands as the most astute version of Llama 2 and is the favorite among users. We recommend to use this variant in your chat application(s) due to its prowess in handling dialogues, logical reasoning, coding.

Downloads

All three model sizes are available on HuggingFace for download:

Model	Llama2	Llama2-hf	Llama2-chat	Llama2-chat-hf
7B	Link	Link	Link	Link
13B	Link	Link	Link	Link
70B	Link	Link	Link	Link

Llama 2 models download (7B, 13B, 70B)