{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 4.784651279449463, "token_generation_latency_sync": 48.0795394897461, "token_generation_latency_async": 47.998166643083096, "token_generation_throughput_sync": 0.020798868096755994, "token_generation_throughput_async": 0.020834129091555784, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 144.3867645263672, "inference_latency_async": 139.62254524230957, "inference_throughput_sync": 0.006925842567913384, "inference_throughput_async": 0.007162167100338548, "inference_CO2_emissions": null, "inference_energy_consumption": null }