Running warmup.. Timing with prompt_tokens=1024, generation_tokens=512, batch_size=1. Trial 1: prompt_tps=3409.194, generation_tps=80.053, peak_memory=5.078, total_time=6.706 Trial 2: prompt_tps=3468.612, generation_tps=81.251, peak_memory=5.079, total_time=6.606 Trial 3: prompt_tps=3304.964, generation_tps=80.468, peak_memory=5.080, total_time=6.683 Trial 4: prompt_tps=3293.373, generation_tps=81.114, peak_memory=5.080, total_time=6.632 Trial 5: prompt_tps=3289.509, generation_tps=79.399, peak_memory=5.081, total_time=6.770 Averages: prompt_tps=3353.131, generation_tps=80.457, peak_memory=5.080