{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 269957.9375, "token_generation_latency_sync": 177.12184295654296, "token_generation_latency_async": 177.28329431265593, "token_generation_throughput_sync": 0.005645831046627891, "token_generation_throughput_async": 0.005640689405491332, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 274.17344055175784, "inference_latency_async": 236.328387260437, "inference_throughput_sync": 0.0036473262982277175, "inference_throughput_async": 0.004231400262965391, "inference_CO2_emissions": null, "inference_energy_consumption": null }