{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 77119.09375, "token_generation_latency_sync": 172.39933319091796, "token_generation_latency_async": 172.3777949810028, "token_generation_throughput_sync": 0.005800486472256728, "token_generation_throughput_async": 0.005801211229730643, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 261.9583480834961, "inference_latency_async": 226.87501907348633, "inference_throughput_sync": 0.003817400771214445, "inference_throughput_async": 0.004407713128064105, "inference_CO2_emissions": null, "inference_energy_consumption": null }