llama.cpp

a2b30183 - speculative : do not apply the MoE cache to the draft model (#31497)