{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 14.32209587097168, "token_generation_latency_sync": 17.65162124633789, "token_generation_latency_async": 17.258471623063087, "token_generation_throughput_sync": 0.05665202000680056, "token_generation_throughput_async": 0.057942558404978674, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 187.64073181152344, "inference_latency_async": 46.34366035461426, "inference_throughput_sync": 0.005329333297444472, "inference_throughput_async": 0.021577924409685388, "inference_CO2_emissions": null, "inference_energy_consumption": null }