{ | |
"current_gpu_type": "Tesla T4", | |
"current_gpu_total_memory": 15095.0625, | |
"perplexity": 2.2062017917633057, | |
"memory_inference_first": 7296.0, | |
"memory_inference": 7296.0, | |
"token_generation_latency_sync": 36.912679481506345, | |
"token_generation_latency_async": 36.8165610358119, | |
"token_generation_throughput_sync": 0.02709096207716405, | |
"token_generation_throughput_async": 0.027161689518673086, | |
"token_generation_CO2_emissions": 0.00017793332995351462, | |
"token_generation_energy_consumption": 0.0022603336588687913, | |
"inference_latency_sync": 1028.5993484497071, | |
"inference_latency_async": 132.02483654022217, | |
"inference_throughput_sync": 0.0009721958326215045, | |
"inference_throughput_async": 0.007574332422637341, | |
"inference_CO2_emissions": 0.00017765837549938682, | |
"inference_energy_consumption": 0.0006121215726133666 | |
} |