DeepSeek V4 Flash runs locally on AMD Ryzen AI MAX+ 395 at up to 32 tok/s

Community build fits DeepSeek V4 Flash plus its speculative draft on a single AMD Ryzen AI MAX+ 395 with 128 GB unified memory, reaching a usable decode rate of up to 32 tokens/s. Details are shared in a linked blog post aimed at Strix Halo owners.