{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 26.50621795654297, "token_generation_latency_sync": 58.16511192321777, "token_generation_latency_async": 58.18860549479723, "token_generation_throughput_sync": 0.01719243661595758, "token_generation_throughput_async": 0.01718549519268703, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 181.9189239501953, "inference_latency_async": 103.4461259841919, "inference_throughput_sync": 0.005496954238107598, "inference_throughput_async": 0.009666867564985613, "inference_CO2_emissions": null, "inference_energy_consumption": null }