{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 221021.65625, "token_generation_latency_sync": 282.71240997314453, "token_generation_latency_async": 283.1099685281515, "token_generation_throughput_sync": 0.0035371634379084815, "token_generation_throughput_async": 0.0035321963588878834, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 272.6269927978516, "inference_latency_async": 234.52019691467285, "inference_throughput_sync": 0.003668015370515727, "inference_throughput_async": 0.004264025073984724, "inference_CO2_emissions": null, "inference_energy_consumption": null }