{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 21.661033630371094, "token_generation_latency_sync": 152.41036834716797, "token_generation_latency_async": 152.54830736666918, "token_generation_throughput_sync": 0.006561233404555194, "token_generation_throughput_async": 0.006555300529139096, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 239.9690734863281, "inference_latency_async": 180.43251037597656, "inference_throughput_sync": 0.0041672036545033105, "inference_throughput_async": 0.005542238468645413, "inference_CO2_emissions": null, "inference_energy_consumption": null }