{ "current_gpu_type": "Tesla T4", "current_gpu_total_memory": 15095.0625, "perplexity": 3.472892999649048, "memory_inference_first": 402.0, "memory_inference": 402.0, "token_generation_latency_sync": 58.756526947021484, "token_generation_latency_async": 58.4649283438921, "token_generation_throughput_sync": 0.01701938579354191, "token_generation_throughput_async": 0.017104271369631656, "token_generation_CO2_emissions": 2.0461894299703792e-05, "token_generation_energy_consumption": 0.0028872648917871455, "inference_latency_sync": 131.08970794677734, "inference_latency_async": 64.51253890991211, "inference_throughput_sync": 0.007628363932323366, "inference_throughput_async": 0.015500862574893231, "inference_CO2_emissions": 1.9946509275385296e-05, "inference_energy_consumption": 6.95395721998155e-05 }