llama.cpp

context : use n_embd_out for pooled embedding extraction

#20840

Merged