{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 7.984891414642334, "token_generation_latency_sync": 22.21034393310547, "token_generation_latency_async": 22.44959268718958, "token_generation_throughput_sync": 0.045024066399505736, "token_generation_throughput_async": 0.04454423801509016, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 336.52049865722654, "inference_latency_async": 146.97182178497314, "inference_throughput_sync": 0.0029715871811380537, "inference_throughput_async": 0.00680402534210298, "inference_CO2_emissions": null, "inference_energy_consumption": null }