Qwen3.8 125B (6B active) Flash-Next
on NVIDIA RTX 3090 · llama.cpp
Summary
User reports Qwen3.8-Flash-next at 160 t/s prefill and 16 t/s decode.
Setup uses a kvarn5 KV cache quant, with experts offloaded to host RAM and n-grams on disk. MTP works but slows decode.
User also mentions alternative setups for 16 GB VRAM with kvarn4 and vision offloaded, and for 12 GB with ub 512.