DGX Spark now runs DeepSeek V4 Flash 284B with a single install command
DGX Spark users can run DeepSeek V4 Flash 284B at 1,000 tokens per second prefill and 59 tokens per second multi-agent serving with one command.
Speculative Decoding News and Analysis
Red Hat AI Releases DFlash Checkpoints to Accelerate Inference for NVIDIA Nemotron Models
Red Hat AI has released DFlash speculator checkpoints for NVIDIA's Nemotron Ultra 550B and Super 120B models, boosting inference speed via speculative decoding.