Llama 2 正式登場！已在 Hugging Face 開源上架並全面支援生態系
Original: Llama 2 is here - get it on Hugging Face
Meta and Microsoft jointly announced Llama 2, a new generation of open-source large language models. Compared to the original Llama, Llama…
Meta 正式發布開源大型語言模型 Llama 2，包含 7B、13B 和 70B 三種參數規模，並允許商用。Hugging Face 與 Meta 深度合作，第一時間在 Hub 上架所有 Llama 2 模型，並全面支援 Transformers、TGI、PEFT 等工具。開發者可以直接在 Hugging Face 平台上進行 Llama 2 的推論、微調（Fine-tuning）與部署，並透過 Spaces 體驗 Chat 版本。
Meta and Microsoft jointly announced Llama 2, a new generation of open-source large language models. Compared to the original Llama, Llama 2 increases training data volume by 40%, doubles the context length to 4,096 tokens, and employs Grouped-Query Attention (GQA) in the 70B model to improve inference efficiency. This release includes three parameter sizes — 7B (7 billion), 13B (13 billion), and 70B (70 billion) — each available in both a base model variant and a dialogue-tuned variant aligned via reinforcement learning from human feedback (RLHF) (Llama-2-chat). Most importantly, Llama 2 is released under a more permissive license that allows free commercial use.
Free shows the 3-line summary; Pro unlocks the full deep summary (~300 words) so you never have to click through.
See Pro plans →Want the original English / full article?
Read on Hugging Face Blog →Summaries are AI-generated; the original article is authoritative.