llama.cpp
ggml-metal: round up to 16 to fix setThreadgroupMemoryLength assertion
#3938
Merged
Merged