{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 7.236781120300293, "token_generation_latency_sync": 58.57762107849121, "token_generation_latency_async": 57.32953827828169, "token_generation_throughput_sync": 0.01707136584908506, "token_generation_throughput_async": 0.017443015067484555, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 181.91411285400392, "inference_latency_async": 103.79562377929688, "inference_throughput_sync": 0.005497099616468762, "inference_throughput_async": 0.009634317552022462, "inference_CO2_emissions": null, "inference_energy_consumption": null }