“The rest of 3.2 should be enlightening as well”
Llama 2: Open Foundation and Fine-Tuned Chat Models
Meta 发布 7B 至 70B 的 Llama 2 及对话模型 Llama 2-Chat 的技术报告
论文发布了 Llama 2,一组参数规模从 70 亿到 700 亿的预训练和微调大语言模型,其中针对对话优化的版本称为 Llama 2-Chat。作者称这些模型在其测试的大多数基准上优于开源对话模型,并且基于人工评估的有用性和安全性,可能适合作为闭源模型的替代品。报告详细描述了微调与安全改进的方法,以便社区在此基础上继续工作。
评论者既把它当作学习 RLHF 和对话模型训练方法的教材,也用它说明许可条款与开源定义的冲突,以及报告没有披露训练数据来源的问题。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“probably the best chat model we can train on, focus on training method.”
“Then, you should check out papers like [OTHER_PAPER] and Llama 2”
“LLaMA 2 and 3 are significantly more capable models trained on orders of magnitude more data, and those papers notably do not say where the data comes from.”
“despite being in direct violation of the terms [1, 2]”
“e.g. Facebook tuned their LLAMA-2 on it [3]”
看全部 7 条 →
“While made by Microsoft, it's widely used in the industry, e.g. Facebook tuned their LLAMA-2 on it [4].”
- Llama 2: Open Foundation and Fine-Tuned Chat Models ↗saeedesmaili · 2023-07-29 · 1 分