Algomaster

Speculative Decoding on a MacBook: How MTP Landed in llama.cpp

Multi-token prediction has moved from research demos to real local inference on Apple Silicon — and the speedup you actually get depends…