llama.cpp

llama-batch: enable parallel sequences for partial rollback

#25278

Merged