Term
Llama 2
Llama 2 is Metas open-weights language model from July 2023 — the first Llama with a commercially usable license, in 7B, 13B and 70B sizes with a 4K context.
Llama 2 — explained in more detail
Llama 2 is the second generation of Metas open-weights language models and was released on July 18, 2023. It was the first Llama released under a license permitting commercial use and contributed substantially to the spread of openly available LLMs. Meta shipped both pretrained base models and RLHF-tuned chat variants (Llama-2-Chat).
Key facts
- Release: July 18, 2023, access via open weights (Hugging Face, Meta).
- Sizes: 7B, 13B and 70B parameters, each as a base and a chat variant.
- Training: pretrained on roughly 2 trillion tokens; the 70B model uses grouped-query attention for faster inference.
- Context window: 4,096 tokens.
- License: Llama 2 Community License — commercial use permitted, with restrictions (including a special clause for very large providers), which is why the term “open source” in the strict sense is contested.
Example / Practical use
Llama 2 could be freely downloaded, self-hosted and fine-tuned, making it the basis of numerous derived models and applications — from chatbots to domain-specific assistants to research prototypes. For operators with data-protection or cost requirements, on-premise capability was the decisive advantage over purely proprietary APIs.
Delimitation
Llama 2 is the successor to the original, research-restricted Llama and the predecessor of Llama 3 and later generations. Against proprietary models such as GPT or Claude it offers open weights, but trails current models in performance and context length. The community license distinguishes it from true open-source models under Apache or MIT licenses.