{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 9842.806640625, "token_generation_latency_sync": 78.03897476196289, "token_generation_latency_async": 78.8802471011877, "token_generation_throughput_sync": 0.012814109911749015, "token_generation_throughput_async": 0.012677445073380644, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 77.85246810913085, "inference_latency_async": 77.07433700561523, "inference_throughput_sync": 0.012844807933362306, "inference_throughput_async": 0.01297448721391071, "inference_CO2_emissions": null, "inference_energy_consumption": null }