AMD ROCm

Run Claude Code Fully Offline Using Qwen3.5 27B and llama.cpp

A developer runs Claude Code CLI against a local llama.cpp server using Qwen3.5 27B, achieving 9+ t/s on Strix Halo hardware.