{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 314.1881103515625, "token_generation_latency_sync": 58.484791946411136, "token_generation_latency_async": 58.827887289226055, "token_generation_throughput_sync": 0.017098462125269887, "token_generation_throughput_async": 0.016998740666710012, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 181.00572357177734, "inference_latency_async": 103.63984107971191, "inference_throughput_sync": 0.005524687177107151, "inference_throughput_async": 0.009648799048532657, "inference_CO2_emissions": null, "inference_energy_consumption": null }