{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 8931.294921875, "token_generation_latency_sync": 262.2849822998047, "token_generation_latency_async": 261.4885849878192, "token_generation_throughput_sync": 0.0038126468058966127, "token_generation_throughput_async": 0.0038242587149514866, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 248.5207061767578, "inference_latency_async": 198.61629009246826, "inference_throughput_sync": 0.004023809586669854, "inference_throughput_async": 0.005034833746690353, "inference_CO2_emissions": null, "inference_energy_consumption": null }