Atomic Chat

Ling 3.1 Flash 560B MoE deployed across four DGX Sparks via llama.cpp

A developer loaded Ling 3.1 Flash, a 560B MoE with 25B active parameters per token, across four DGX Sparks using llama.cpp RPC and Atomic Chat's day-one Q4 GGUF, since no official vLLM FP4 build was available. He expects bandwidth and speed to be painful and wants to compare it with the 124B Ling 3.0 Flash.