Meta 在 Hugging Face 开源 816B 参数模型 Muse Spark Open
Meta released the weights of Muse Spark Open, on Huggingface, an 816B-parameter model built for coding and agentic work.
Meta 在 Hugging Face 发布 Muse Spark Open 的模型权重，这是一个 816B 参数、面向编码与智能体任务的模型。开源版与 Muse Spark 1.3 共用同一基础模型，额外做了安全后训练，以降低其脱离系统护栏单独使用时的滥用风险，这也逆转了 Meta 今年 4 月首个专有闭源前沿模型的策略。模型共 62 层，每层含 256 个专家网络，每个 token 路由到其中 8 个，因此部署需要 816B 参数模型的内存，但运行速度相当于 42B 模型。一次在 512GB M3 Ultra 上测试的 3-bit 量化版本约占 344 GB，短提示词下生成速度约 21 tok/s。
推荐理由
精选AI 评分 82
Meta 把 816B 参数的 MoE 模型权重放到 Hugging Face，并给出 3-bit 量化在 512GB M3 Ultra 上的实测速度，读者可据此判断本地部署门槛。
主题
查看原帖Muse Spark Open -- 权重今日发布 -- 得益于与 @johnternus 和 Apple 的合作，在新款 Mac Studios 上运行出色。https://t.co/U2wJNrKBOg
查看原文
Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple. https://t.co/U2wJNrKBOg
正文
Meta released the weights of Muse Spark Open, on Huggingface, an 816B-parameter model built for coding and agentic work.
The open version shares the same base model as Muse Spark 1.3, with additional safety post-training to resist misuse as a standalone model without system guardrails, reversing Meta’s first proprietary, closed-weight frontier model strategy from April.
Each of its 62 layers holds 256 expert networks and routes every token through 8, so deployment needs the memory of an 816B-parameter model but runs at the speed of a 42B model.
That storage bill explains the Apple tie-up, because the M5 Ultra pushes 1.2 TB/s of unified memory and a Mac Studio can carry up to 512GB.
A 3-bit quant tested on a 512GB M3 Ultra took about 344 GB and ran at about 21 tok/s generation on a short prompt.