{ | |
"current_gpu_type": "Tesla T4", | |
"current_gpu_total_memory": 15095.0625, | |
"perplexity": 30815030.0, | |
"memory_inference_first": 1142.0, | |
"memory_inference": 1142.0, | |
"token_generation_latency_sync": 147.6204864501953, | |
"token_generation_latency_async": 147.56362289190292, | |
"token_generation_throughput_sync": 0.006774127521503482, | |
"token_generation_throughput_async": 0.006776737927697435, | |
"token_generation_CO2_emissions": 0.00020048614177282614, | |
"token_generation_energy_consumption": 0.008781923951582248, | |
"inference_latency_sync": 1151.6722137451172, | |
"inference_latency_async": 424.6630907058716, | |
"inference_throughput_sync": 0.0008683026194997837, | |
"inference_throughput_async": 0.00235480789803938, | |
"inference_CO2_emissions": 0.0002004475962713345, | |
"inference_energy_consumption": 0.0006890801651842688 | |
} |