Roughly 8-12 token/s on generation depending on context length. Prompt processing is faster obviously. Haven't benchmarked it super carefully though, just eyeba

Roughly 8-12 token/s on generation depending on context length. Prompt processing is faster obviously. Haven't benchmarked it super carefully though, just eyeballing the llama.cpp output.