|
{ |
|
"current_gpu_type": "Tesla T4", |
|
"current_gpu_total_memory": 15095.0625, |
|
"perplexity": 3.4586403369903564, |
|
"memory_inference_first": 808.0, |
|
"memory_inference": 808.0, |
|
"token_generation_latency_sync": 39.485802841186526, |
|
"token_generation_latency_async": 38.54773212224245, |
|
"token_generation_throughput_sync": 0.025325558252469627, |
|
"token_generation_throughput_async": 0.025941863371593512, |
|
"token_generation_CO2_emissions": 1.9341882246324658e-05, |
|
"token_generation_energy_consumption": 0.0018858868968200813, |
|
"inference_latency_sync": 122.10734748840332, |
|
"inference_latency_async": 48.58226776123047, |
|
"inference_throughput_sync": 0.008189515377811079, |
|
"inference_throughput_async": 0.02058364185292351, |
|
"inference_CO2_emissions": 1.933046505752413e-05, |
|
"inference_energy_consumption": 6.813219116923538e-05 |
|
} |