How one Redditor squeezed Raon-1-mini from 50 to 232 tok/s on a 16GB RTX 5070 Ti

McNiiby · reddit · 2026-10-29

A Reddit user with only an RTX 5070 Ti (16GB VRAM) documented pushing KRAFTON's Raon-1-mini-preview (29B MoE, 4B active) decode speed from 50 to 232 tok/s at roughly Q4KM quality.

Key steps:

The author released the IQ3KS GGUF on Hugging Face and a fork of ikllama.cpp with build/run instructions (needs 16GB card) — an excellent engineering writeup on running MoE models locally on consumer GPUs.

Original post →

More from coding & agent

coding & agent channel →