Everything on DGX Station
1 mention, newest first.
NVIDIA tutorial serves GLM-5.3 NVFP4 across two DGX Stations with vLLM
NVIDIA's tutorial serves the GLM-5.3 NVFP4 model, a quantized version of Z.ai's MoE model with 753 billion parameters (40 billion active per token), across two DGX Stations with vLLM. The checkpoint is about 464 GB, so tensor parallelism (TP=2) splits it over two Stations connected by ConnectX-8 RDMA. Station A hosts the API endpoint and Station B runs a worker.
Not yet