{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 172285.734375, "token_generation_latency_sync": 174.6008560180664, "token_generation_latency_async": 175.03785993903875, "token_generation_throughput_sync": 0.005727348781706588, "token_generation_throughput_async": 0.00571304973877237, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 272.2702331542969, "inference_latency_async": 234.3303680419922, "inference_throughput_sync": 0.003672821624364993, "inference_throughput_async": 0.004267479321420258, "inference_CO2_emissions": null, "inference_energy_consumption": null }