Running warmup.. Timing with prompt_tokens=1024, generation_tokens=512, batch_size=1. Trial 1: prompt_tps=1538.942, generation_tps=86.573, peak_memory=20.660, total_time=6.585 Trial 2: prompt_tps=1350.163, generation_tps=87.959, peak_memory=20.660, total_time=6.584 Trial 3: prompt_tps=1371.898, generation_tps=86.726, peak_memory=20.661, total_time=6.655 Trial 4: prompt_tps=1314.767, generation_tps=84.044, peak_memory=20.661, total_time=6.877 Trial 5: prompt_tps=1291.455, generation_tps=87.038, peak_memory=20.661, total_time=6.679 Averages: prompt_tps=1373.445, generation_tps=86.468, peak_memory=20.661