{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 6.853957176208496, "token_generation_latency_sync": 19.214177703857423, "token_generation_latency_async": 19.127227552235126, "token_generation_throughput_sync": 0.052044902228589304, "token_generation_throughput_async": 0.052281492300390614, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 184.96829681396486, "inference_latency_async": 47.145628929138184, "inference_throughput_sync": 0.005406331880785861, "inference_throughput_async": 0.021210874108881676, "inference_CO2_emissions": null, "inference_energy_consumption": null }