speculative decoding in llama.cpp

[vrambandit]

A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at tail latency. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.