{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 4977.65625, "token_generation_latency_sync": 68.87611618041993, "token_generation_latency_async": 68.9922260120511, "token_generation_throughput_sync": 0.014518820970980935, "token_generation_throughput_async": 0.014494386654886691, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 68.0677375793457, "inference_latency_async": 66.47746562957764, "inference_throughput_sync": 0.014691247800535644, "inference_throughput_async": 0.01504269139218016, "inference_CO2_emissions": null, "inference_energy_consumption": null }