DeepSeek-V4 on Day 0: From Fast Inference to Verified RL with SGLang and Miles

“SGLang’s day-zero support for DeepSeek-V4 covers fast inference and verified reinforcement learning.”

Fast inference for DeepSeek-V4 on commodity GPUs is the actually-interesting piece. The Chinese open-weight ecosystem keeps shipping models that match closed-weights performance and shipping the inference stack to run them. The US labs are losing this race in slow motion.