{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 116018.1484375, "token_generation_latency_sync": 82.89982223510742, "token_generation_latency_async": 82.94181060045958, "token_generation_throughput_sync": 0.012062752042627516, "token_generation_throughput_async": 0.012056645409118414, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 82.54658432006836, "inference_latency_async": 81.7939281463623, "inference_throughput_sync": 0.012114371639201605, "inference_throughput_async": 0.01222584637591432, "inference_CO2_emissions": null, "inference_energy_consumption": null }