vLLM publishes NVIDIA and AMD images to serve Muse Spark Open

vLLM says users can run Muse Spark Open using NVIDIA and AMD container images, with a recipe and model link provided. The accompanying diagram shows the base model and DSpark model feeding one deployment recipe, which selects hardware-specific flags.

The image lists a parser of muse, a maximum context of 1,048,576 tokens, and BF16 and NVFP4 versions, and notes that public availability should be confirmed at launch.