{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 92177.453125, "token_generation_latency_sync": 89.34934310913086, "token_generation_latency_async": 89.74834606051445, "token_generation_throughput_sync": 0.01119202408436965, "token_generation_throughput_async": 0.011142266614314339, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 90.9167610168457, "inference_latency_async": 88.7260913848877, "inference_throughput_sync": 0.01099907199525853, "inference_throughput_async": 0.011270641864094617, "inference_CO2_emissions": null, "inference_energy_consumption": null }