NVIDIA posts NVFP4 checkpoints tuned for DGX Spark as Blackwell runs FP4 natively

AIFlow_ML · x · 2026-11-21

NVIDIA released NVFP4 quantized model checkpoints tuned for DGX Spark. Since Blackwell runs FP4 natively, these checkpoints are a strong starting point for agent workloads on the device, offering better local inference efficiency. A notable new weight option for on-device agent deployment.

Original post →

More from Infra

Infra channel →