{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 7.147127151489258, "token_generation_latency_sync": 290.7360534667969, "token_generation_latency_async": 290.7614739611745, "token_generation_throughput_sync": 0.003439545897647688, "token_generation_throughput_async": 0.0034392451873920906, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 460.26710510253906, "inference_latency_async": 368.649697303772, "inference_throughput_sync": 0.0021726514645821113, "inference_throughput_async": 0.0027126022544268834, "inference_CO2_emissions": null, "inference_energy_consumption": null }