{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 13.287712097167969, "token_generation_latency_sync": 17.040970897674562, "token_generation_latency_async": 16.898038238286972, "token_generation_throughput_sync": 0.05868210244619699, "token_generation_throughput_async": 0.05917846710360944, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 188.50826263427734, "inference_latency_async": 49.17123317718506, "inference_throughput_sync": 0.005304807258979879, "inference_throughput_async": 0.02033709417855295, "inference_CO2_emissions": null, "inference_energy_consumption": null }