{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 9.791707992553711, "token_generation_latency_sync": 18.775645637512206, "token_generation_latency_async": 18.96346565335989, "token_generation_throughput_sync": 0.053260485381236726, "token_generation_throughput_async": 0.05273297709814045, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 190.64278869628907, "inference_latency_async": 47.53875732421875, "inference_throughput_sync": 0.005245412149279294, "inference_throughput_async": 0.021035467822179424, "inference_CO2_emissions": null, "inference_energy_consumption": null }