{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 469.59527587890625, "token_generation_latency_sync": 399.7762176513672, "token_generation_latency_async": 399.4270246475935, "token_generation_throughput_sync": 0.002501399422594142, "token_generation_throughput_async": 0.002503586233010348, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 384.87603454589845, "inference_latency_async": 335.98711490631104, "inference_throughput_sync": 0.0025982391997461324, "inference_throughput_async": 0.002976304612987456, "inference_CO2_emissions": null, "inference_energy_consumption": null }