llama.cpp

Use correct type of pooling for embedding models

#5500

Merged