llama.cpp

faa1bc26 - sampling : delegate input allocation to the scheduler (#19266)

sampling : delegate input allocation to the scheduler (#19266) * sampling : delegate input allocation to the scheduler * graph : compute backend samplers only if needed