pod benchmark: throughput + GPU-util sweep, reasoning on/off 7df674a verified PS4Research commited on 22 days ago
fix: port probe logic, fail-fast on server death, symlink-safe cache 1ce3c17 verified PS4Research commited on 22 days ago
fat llama.cpp bundle: sm_80+sm_120, CUDA 12.8, self-contained 9a1fd71 verified PS4Research commited on 22 days ago