{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 34524.34375, "token_generation_latency_sync": 156.95160369873048, "token_generation_latency_async": 157.1216208860278, "token_generation_throughput_sync": 0.006371390775461625, "token_generation_throughput_async": 0.006364496460518158, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 247.08700256347657, "inference_latency_async": 200.04968643188477, "inference_throughput_sync": 0.004047157436956241, "inference_throughput_async": 0.004998758147718925, "inference_CO2_emissions": null, "inference_energy_consumption": null }