llama.cpp

4efe9898 - context : pass embeddings tensor from encoder to decoder