(圖說:拍攝於 Lac de Neuchatel 湖畔,瑞士。圖片來源:Ernest。)
關於
整理 Llama 1 與 Llama 2 的比較資訊,以及關於 LLAMA 2 COMMUNITY LICENSE AGREEMENT 的筆記備忘。更多 Meta AI 模型與產品請參考這頁筆記。
Llama 1
- Blog = Introducing LLaMA: A foundational, 65-billion-parameter language model on 2023-02-24
- Paper = [2302.13971] LLaMA: Open and Efficient Foundation Language Models
- License = GPL v3
Llama 2
- Blog = Meta and Microsoft Introduce the Next Generation of Llama on 2023-07-18
- Paper = [2307.09288] Llama 2: Open Foundation and Fine-Tuned Chat Models
- License = Llama 2 Community License Agreement
- 填完 Request 表格後會收到一封信說明同意授權、及下載 models 的方法。
LLAMA 2 COMMUNITY LICENSE AGREEMENT
llama/LICENSE at 6d4c0c290aeec1fa4399694fefb864be5a153bb6 · facebookresearch/llama
我將大家討論比較多的段落標示出來。License 全文請參考上面連結。
- You are granted a non-exclusive, worldwide, non-transferable and royalty-free limited license under Meta’s intellectual property or other rights owned by Meta embodied in the Llama Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the Llama Materials.
- 可散佈。
- Licensee 用 Llama Materials 製作出衍生物,該衍生物的終端使用者不適用本授權的 Section 2?(Section 2 可能只管轄機構、不適用終端使用者?)
- 不能拿 Llama 來改進其他 LLM。
- 超過 7 億 MAU 要申請許可。 (Section 2)
![](/en/notes/ai/meta-ai-llama-2/llama-2-license-1_hu13173348291224394499.png)
![](/en/notes/ai/meta-ai-llama-2/llama-2-license-2_hu11299890909499399959.png)
![](/en/notes/ai/meta-ai-llama-2/llama-2-license-3_hu10868413980474533950.png)
Llama 1 vs Llama 2 (Comparison)
討論
- 雖然關於 “open source” 這詞有爭議、有各種詮釋角度?!1 但 Meta AI 這一步,也許可以取得 Edge AI 的先期導入佔比?
- 七億 MAU 以下,請別想太多,盡情創造無腦玩?!
- 期待更多測試、比較結果,各種模型或是工具,應有其優缺點、適用性。
- 同一天 (2023-07-18) AWS Blog 也發布一篇 Llama 2 foundation models from Meta are now available in Amazon SageMaker JumpStart | AWS Machine Learning Blog,推測 Meta 應該也有與 AWS 密切合作,才能在 2023-07-18 同時發表?而 Meta Blog 與新聞稿卻是與 Microsoft 聯名,頗有趣。
Meta launches Llama 2, an open source AI model that allows commercial applications | Ars Technica
Although Llama 2 is open source, Meta did not disclose the source of the training data used in creating the Llama 2 models, which Mozilla Senior Fellow of Trustworthy AI Abeba Birhane pointed out on Twitter. Lack of training data transparency is still a sticking point for some LLM critics because the training data that teaches these LLMs what they "know" often comes from an unauthorized scrape of the Internet with little regard for privacy or commercial impact.
↩︎