llama.cpp

Implement '--keep-split' to quantize model into several shards

#6688

Merged