{ "current_gpu_type": "Tesla T4", "current_gpu_total_memory": 15095.0625, "perplexity": 5.245799541473389, "memory_inference_first": 362.0, "memory_inference": 362.0, "token_generation_latency_sync": 95.34894561767578, "token_generation_latency_async": 95.61750814318657, "token_generation_throughput_sync": 0.010487792953786162, "token_generation_throughput_async": 0.010458335710887872, "token_generation_CO2_emissions": 2.0800614727756692e-05, "token_generation_energy_consumption": 0.004631853474529401, "inference_latency_sync": 146.7707092285156, "inference_latency_async": 94.17455196380615, "inference_throughput_sync": 0.006813348557463488, "inference_throughput_async": 0.010618579851426608, "inference_CO2_emissions": 2.1171634369010966e-05, "inference_energy_consumption": 7.173729223505599e-05 }