跳到正文
NVIDIA AI· @NVIDIAAI · X·· 3 小时前同新闻AI 评分79
AI 导读

Meta 发布开源权重模型 Muse Spark Open,总参数 816B、激活 42B(MoE),在 Artificial Analysis Intelligence Index 得 39 分,为美国开源权重模型中最高。该模型 token 用量较少,每任务成本 $0.84,支持文本、图像、视频、音频输入,1M token 上下文窗口,BF16 权重约 1.6TB,NVFP4 量化后约 530GB,可在四张 Blackwell Ultra GPU、八张 H200 或两台 512GB M5 Ultra Mac Studio 上自托管,权重已上 Hugging Face。其相对短板是智能体评测,在九个高分开源权重模型中的四项智能体评估上排名末位;采用 Muse Model License,要求标注归属,年营收超 $20M 的模型服务商家需与 Meta 另行签约。

同一新闻,精选展示《Meta 发布开源权重多模态 MoE 模型 Muse Spark Open:816B 参数、1M 上下文》

正文 · AI 翻译

恭喜 @alexandr_wang 和 @Muse 团队发布如此史诗级的开放权重模型!

引用Artificial Analysis@ArtificialAnlys
Meta 发布了 Muse Spark Open,这是一个 816B(42B 激活)开放权重模型,在 Artificial Analysis Intelligence Index 上得分 39,成为美国排名第一的开放权重模型 Muse Spark Open(xhigh)比 Muse Glimmer(high,17)高 22 分,后者是 Meta 于 8 月发布的 30B 稠密开放权重模型。它也领先于 NVIDIA 的 Nemotron 3.5 Super VL(26),这是得分第二高的美国开放权重模型,于今天发布 @AIatMeta 在公开发布前与我们分享了访问权限以进行基准测试。祝贺 @AIatMeta、@finkd 和 @alexandr_wang 发布! 要点: ➤ 开放权重达到 Meta 8 月旗舰水平。Muse Spark Open 的得分与 Muse Spark 1.2(xhigh,40)相差不到一分。它在 AutomationBench-AA 上更强,幻觉率更低,但在 GDPval-AA v2.1、CritPt 和 AA-Omniscience 准确率上更弱。在 Meta 当前阵容中,它比 Muse Wonder(max,52)低 13 分,比 Muse Wonder Full(max,56)低 17 分 ➤ Token 高效,每任务成本适中。Muse Spark Open 在每个 Intelligence Index 任务上使用 40k 输出 token,是九个得分最高的开放权重模型中最少的:例如比 Muse Spark 1.2(xhigh,48k)少 16%,比 GLM-5.3(max,71k)少 43%,后者是其按规模最接近的中国同类模型。Together AI、Fireworks 和 Baseten 在发布时以每 1M 输入/输出 token $1.25/$4.25 提供(缓存输入 $0.15)。按此定价,每个 Intelligence Index 任务成本为 $0.84。这低于相同每 token 价格下的 Muse Spark 1.2($0.97),也低于 GLM-5.3(max,$2.01)的一半,但约为 DeepSeek V4.1 Flash(max,$0.27)的三倍 ➤ 与中国开放权重领先者相比,在智能体工作上相对较弱。在九个得分最高的开放权重模型中,Muse Spark Open 在索引中的四项智能体评估上排名最后,但在其他六项合计中排名第四,并且在 AA-Omniscience(25)上领先所有开放权重模型。按规模最接近的中国模型 GLM-5.3(max)其 6 分领先全部来自智能体评估 ➤ 可在单节点或两台 Mac Studio 上自托管。BF16 权重约为 1.6TB。Meta 的 NVFP4 checkpoint 仅量化 256 个专家,将权重降至约 530GB。Meta 表示这可在四个 Blackwell Ultra GPU 或八个 H200 上运行,也可使用 MLX 在两台 512GB M5 Ultra Mac Studio 上运行。该架构使用 62 层,采用分组查询注意力(16 个 KV 头),每个 token 激活 256 个专家中的 8 个,并包含一个 1.8B 参数的 Perception Encoder 用于视觉 其他模型细节: ➤ 参数:总计 816B,激活 42B(MoE) ➤ 上下文窗口:1M token ➤ 许可证:Muse Model License,这是一种自定义许可证,要求标注“Built with Muse Spark Open”,并且年收入超过 $20M 的 Model-as-a-Service 业务需与 Meta 另行达成协议 ➤ 输入模态:文本、图像、视频和音频。输出:文本 ➤ 可用性:权重在 Hugging Face 上(BF16 和 NVFP4),API 访问可通过 Baseten、Fireworks AI、Groq、Modal、Nebius、OpenRouter 和 Together AI
原文

Meta has released Muse Spark Open, an 816B (42B active) open-weights model that scores 39 on the Artificial Analysis Intelligence Index, making it the #1 US open-weights model Muse Spark Open (xhigh) scores 22 points above Muse Glimmer (high, 17), Meta's 30B dense open-weights model from August. It is also ahead of NVIDIA's Nemotron 3.5 Super VL (26), the next highest-scoring US open-weights model, which launched today @AIatMeta shared access with us ahead of public release for benchmarking. Congratulations to @AIatMeta, @finkd, and @alexandr_wang on the release! Key Takeaways: ➤ Open weights at the level of Meta's August flagship. Muse Spark Open scores within a point of Muse Spark 1.2 (xhigh, 40). It is stronger on AutomationBench-AA and has a lower hallucination rate, but weaker on GDPval-AA v2.1, CritPt and AA-Omniscience accuracy. Within Meta's current lineup, it sits 13 points behind Muse Wonder (max, 52) and 17 points behind Muse Wonder Full (max, 56) ➤ Token-efficient, at a moderate cost per task. Muse Spark Open uses 40k output tokens per Intelligence Index task, the fewest among the nine highest-scoring open-weights models: e.g. 16% fewer than Muse Spark 1.2 (xhigh, 48k) and 43% fewer than GLM-5.3 (max, 71k), its closest Chinese peer by size. Together AI, Fireworks and Baseten offer it at launch for $1.25/$4.25 per 1M input/output tokens ($0.15 cached input). At that pricing it costs $0.84 per Intelligence Index task. That is less than Muse Spark 1.2 ($0.97) at the same per-token prices and less than half of GLM-5.3 (max, $2.01), but about three times DeepSeek V4.1 Flash (max, $0.27) ➤ A relative weakness in agentic work compared to Chinese open-weights leaders. Among the nine highest-scoring open-weights models, Muse Spark Open ranks last on the four agentic evaluations in the index but fourth on the other six combined, and it leads all open-weights models on AA-Omniscience (25). GLM-5.3 (max), the closest Chinese model by size, owes all of its 6-point lead to the agentic evaluations ➤ Self-hostable on a single node, or two Mac Studios. BF16 weights are ~1.6TB. Meta's NVFP4 checkpoint quantizes only the 256 experts, cutting the weights to ~530GB. Meta says this runs on four Blackwell Ultra GPUs or eight H200s, or on two 512GB M5 Ultra Mac Studios using MLX. The architecture uses 62 layers with grouped-query attention (16 KV heads), 8 of 256 experts active per token, and a 1.8B-parameter Perception Encoder for vision Other model details: ➤ Parameters: 816B total, 42B active (MoE) ➤ Context window: 1M tokens ➤ License: Muse Model License, a custom license that requires "Built with Muse Spark Open" attribution and a separate agreement with Meta for Model-as-a-Service businesses with over $20M in annual revenue ➤ Input modalities: text, image, video and audio. Output: text ➤ Availability: weights on Hugging Face (BF16 and NVFP4), with API access via Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI

在 X 查看被引用的帖子

来源:NVIDIA AI · x.com