Qwen 3.6 27B MTP on v100 32GB: 54 t/s
Just a quick note that I got a nice result using am17an's MTP branch of llama.cpp on v100 32GB SXM module using one of those pcie card adapters. Pulled and built in one shot, and llama-server ran without a hitch. Tested using am17an's MTP GGUF,…