{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 9.83940315246582, "token_generation_latency_sync": 50.838589096069335, "token_generation_latency_async": 51.20471827685833, "token_generation_throughput_sync": 0.01967009741577027, "token_generation_throughput_async": 0.019529450286067566, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 190.01671752929687, "inference_latency_async": 103.03716659545898, "inference_throughput_sync": 0.0052626948460249, "inference_throughput_async": 0.009705235819674331, "inference_CO2_emissions": null, "inference_energy_consumption": null }