{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 224556.421875, "token_generation_latency_sync": 163.17259521484374, "token_generation_latency_async": 163.60859014093876, "token_generation_throughput_sync": 0.006128480083823723, "token_generation_throughput_async": 0.006112148507230344, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 251.73944244384765, "inference_latency_async": 219.0598726272583, "inference_throughput_sync": 0.003972361225130851, "inference_throughput_async": 0.0045649620261651105, "inference_CO2_emissions": null, "inference_energy_consumption": null }