{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": Infinity, "token_generation_latency_sync": 9.927025508880615, "token_generation_latency_async": 9.78469680994749, "token_generation_throughput_sync": 0.10073510933415153, "token_generation_throughput_async": 0.10220040737321186, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 54.19745292663574, "inference_latency_async": 17.0029878616333, "inference_throughput_sync": 0.01845105159007468, "inference_throughput_async": 0.05881319260695751, "inference_CO2_emissions": null, "inference_energy_consumption": null }