Meta 蒸馏出 Muse Glimmer 30B，单张消费级 GPU 跑本地 Agent
Meta 把自家大模型蒸馏成 30B 小模型，一张 RTX 5090 就能跑 Agent，MCPAtlas 上比分(open-source)同级模型高不少，玩本地部署的可以看看。
Meta 从更大的 Muse Spark 蒸馏出 30B 稠密模型 Muse Glimmer，并用模拟应用中的合成 Agent 运行数据训练。其 4-bit 权重不到 20 GB，可在一块 RTX 5090 上运行。在多工具基准 MCPAtlas 上得 75.5 分，超过 Gemma4-31B 的 54.2 分和 Qwen3.6-27B 的 62.5 分。但在 OSWorld-Verified 上为 65.9 分，低于 Qwen3.6-27B 的 75.6 分。
Meta's paper behind, Muse Glimmer 30B, a 30B model distilled from Meta's larger Muse Spark can run AI agents on 1 consumer GPU
Open models small enough for consumer hardware are usually general-purpose and weak at long tool-using tasks. That leaves local agents well behind cloud-hosted ones.
Meta distilled Muse Glimmer, a dense 30B model, from its larger Muse Spark and trained it on synthetic agent runs in mock apps. Its 4-bit weights fit in under 20 GB, so it runs on a single RTX 5090.
On MCPAtlas, a multi-tool test, it scored 75.5 against 54.2 for Gemma4-31B and 62.5 for Qwen3.6-27B. It still trails Qwen3.6-27B on OSWorld-Verified, 65.9 against 75.6.
