base_results.json

{
    "current_gpu_type": "Tesla T4",
    "current_gpu_total_memory": 15095.0625,
    "perplexity": 2.2062017917633057,
    "memory_inference_first": 7298.0,
    "memory_inference": 7298.0,
    "token_generation_latency_sync": 35.75132808685303,
    "token_generation_latency_async": 35.67290659993887,
    "token_generation_throughput_sync": 0.027970988869857785,
    "token_generation_throughput_async": 0.02803247885614439,
    "token_generation_CO2_emissions": 0.00018359090889950798,
    "token_generation_energy_consumption": 0.002221255088823526,
    "inference_latency_sync": 1058.6026702880858,
    "inference_latency_async": 136.83900833129883,
    "inference_throughput_sync": 0.000944641486430279,
    "inference_throughput_async": 0.007307857694926547,
    "inference_CO2_emissions": 0.00018326470280637688,
    "inference_energy_consumption": 0.000631884688384694
}