{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 20459.171875, "token_generation_latency_sync": 167.92886505126953, "token_generation_latency_async": 168.31964384764433, "token_generation_throughput_sync": 0.005954902390930201, "token_generation_throughput_async": 0.005941077209652112, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 266.4105987548828, "inference_latency_async": 212.85629272460938, "inference_throughput_sync": 0.0037536044161669145, "inference_throughput_async": 0.0046980053405975015, "inference_CO2_emissions": null, "inference_energy_consumption": null }