Running warmup.. Timing with prompt_tokens=1024, generation_tokens=512, batch_size=1. Trial 1: prompt_tps=466.746, generation_tps=17.777, peak_memory=16.991, total_time=31.008 Trial 2: prompt_tps=469.500, generation_tps=17.784, peak_memory=16.991, total_time=30.986 Trial 3: prompt_tps=468.553, generation_tps=17.600, peak_memory=16.992, total_time=31.291 Trial 4: prompt_tps=447.236, generation_tps=17.379, peak_memory=16.992, total_time=31.765 Trial 5: prompt_tps=444.815, generation_tps=17.255, peak_memory=16.993, total_time=31.988 Averages: prompt_tps=459.370, generation_tps=17.559, peak_memory=16.992