{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 9.284024238586426, "token_generation_latency_sync": 58.49337272644043, "token_generation_latency_async": 58.878608234226704, "token_generation_throughput_sync": 0.01709595383868121, "token_generation_throughput_async": 0.01698409711081945, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 94.05050811767578, "inference_latency_async": 32.421159744262695, "inference_throughput_sync": 0.010632584767631476, "inference_throughput_async": 0.030844053941560857, "inference_CO2_emissions": null, "inference_energy_consumption": null }