2 ms·M2 Ultra runs codellama 34B F16 at 150 token per second3 points by behnamoh 3y agosp332 3y agoThat's the "prompt processing" speed, just reading the prompt. It can only generate text at 28 t/s according to that chart.behnamoh 3y agono, prompt processing is 512 (look at pp).sp332 3y ago512 means the prompt had 512 tokens total. The next column, labelled "t/s", shows tokens per second. The text generation benchmark is labelled "tg 64" because it generates 64 tokens for the benchmark.