{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 13.530919075012207, "token_generation_latency_sync": 255.42847747802733, "token_generation_latency_async": 255.30402418226004, "token_generation_throughput_sync": 0.003914990254311103, "token_generation_throughput_async": 0.003916898698338205, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 245.6321014404297, "inference_latency_async": 184.9421262741089, "inference_throughput_sync": 0.004071129116006518, "inference_throughput_async": 0.005407096912673464, "inference_CO2_emissions": null, "inference_energy_consumption": null }