{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 9.573497772216797, "token_generation_latency_sync": 55.56139488220215, "token_generation_latency_async": 54.98296096920967, "token_generation_throughput_sync": 0.017998108257003598, "token_generation_throughput_async": 0.018187452664835523, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 194.78456268310546, "inference_latency_async": 109.46733951568604, "inference_throughput_sync": 0.005133877070262994, "inference_throughput_async": 0.009135144824239615, "inference_CO2_emissions": null, "inference_energy_consumption": null }