Muse Spark Open 816B-A42B

Muse Spark Open 816B‑A42B is an open‑weight multimodal mixture‑of‑experts model with ~816 billion total parameters and ~42 billion active per token. It supports text, images, audio and video, a 1 M‑token context window, and is trained for agentic coding tasks such as multi‑file refactoring and long debugging. The architecture contains 62 layers, hidden dimension 8192, 256 experts per MoE layer with a router selecting 8 experts per token. Checkpoints are available in BF16 and NVFP4 formats. The BF16 checkpoint is ~1.5 TB and runs on 4 nodes of 8 H200 GPUs (TP = 32). The NVFP4 variant uses 4‑bit weights and runs on a single NVIDIA Blackwell chassis. Muse-Spark-Open-816B-A42B has model ID meta-vlm-muse-spark-open-816b-a42b Muse-Spark-Open-816B-A42B runs on instance type ml.p6-b300.48xlarge Amazon SageMaker JumpStart contains the model Muse-Spark-Open-816B-A42B Granite‑Switch‑4.1‑30B model outperforms Jev‑27B model on real‑world 5‑step RAG workflow Mark Zuckerberg collaborated with Alexandr Wang· Graphiti 生成的实体摘要