{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 9.484892845153809, "token_generation_latency_sync": 164.63221588134766, "token_generation_latency_async": 164.83410950750113, "token_generation_throughput_sync": 0.006074145298030317, "token_generation_throughput_async": 0.006066705507663709, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 260.58793182373046, "inference_latency_async": 192.69790649414062, "inference_throughput_sync": 0.0038374762522634014, "inference_throughput_async": 0.005189469975017124, "inference_CO2_emissions": null, "inference_energy_consumption": null }