Llama 3 上线 Ollama，提供 8B 与 70B 两个版本
Llama 3 · Ollama Blog
Llama 3 现已可通过 Ollama 运行，首发包含 8B 和 70B 两个参数规模，命令分别为 ollama run llama3:8b 和 ollama run llama3:70b。相比 Llama 2，其训练数据集扩大七倍，上下文长度翻倍至 8K，token 词表扩大到 128K，误拒答情况不到 Llama 2 的三分之一。Meta 计划后续发布 400B 参数版本及支持多模态、多语言和更长上下文窗口的多个模型。
推荐理由
AI 评分 76
原文列出 Llama 3 相对 Llama 2 的训练数据、上下文与 token 词表变化，并给出 Ollama 上的运行命令。
主题
正文
当前显示原文，中文全文译文尚未就绪。
April 18, 2024
Llama 3 is now available to run using Ollama.
The most capable model
Llama 3 represents a large improvement over Llama 2 and other openly available models:
Trained on a dataset seven times larger than Llama 2; Double the context length of 8K from Llama 2; Encodes language much more efficiently using a larger token vocabulary with 128K tokens; Less than 1⁄3 of the false “refusals” when compared to Llama 2
Two sizes: 8B and 70B parameters
The initial release of Llama 3 includes two sizes:
8B Parameters ollama run llama3:8b; 70B Parameters ollama run llama3:70b
Using Llama 3 with popular tooling
Meta plans to release a 400B parameter Llama 3 model and many more. Over the coming months, they will release multiple models with new capabilities including multimodality, the ability to converse in multiple languages, a much longer context window, and stronger overall capabilities.