{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 5.671560764312744, "token_generation_latency_sync": 61.94393539428711, "token_generation_latency_async": 61.734722182154655, "token_generation_throughput_sync": 0.016143630423781998, "token_generation_throughput_async": 0.016198339680696984, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 326.5973266601562, "inference_latency_async": 214.43421840667725, "inference_throughput_sync": 0.003061874419567919, "inference_throughput_async": 0.004663434816655461, "inference_CO2_emissions": null, "inference_energy_consumption": null }