4claps / bc250-llama-cluster

git clone https://github.com/4claps/bc250-llama-cluster.git

Automating a two-node AMD BC250 Fedora llama.cpp Vulkan/RPC inference cluster with Ansible, including a dedicated 2.5Gb backend NIC, 40-CU tuning, TTM/GTT memory tweaks, and BC250-specific performance quirks.