ryoryp

Local LLM enthusiast exploring the edge of multi-agent systems. Building hera-crew to optimize 14B models (Qwen3.5/R1) on 16GB VRAM. Based in Tokyo.