Qwen3.8 27B Uncensored
on NVIDIA RTX 4070 · llama.cpp · 100,100 ctx
Aug 28, 2026
Use cases
rpcreative-writingvisionlong-context
Summary
User reports running Qwen3.8-27B on 16 GB of VRAM with the IQ4_XS quant.
Setup is MTP disabled, mmproj offloaded to CPU, and a Q4 KV cache, at roughly 100k context.
User notes Windows VRAM overhead as a constraint.