base_results.json

{
    "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": 36.71660461425781,
    "token_generation_latency_async": 36.8287518620491,
    "token_generation_throughput_sync": 0.027235633863913426,
    "token_generation_throughput_async": 0.027152698623774684,
    "token_generation_CO2_emissions": 1.9246105981369177e-05,
    "token_generation_energy_consumption": 0.0018056219463504367,
    "inference_latency_sync": 123.11312103271484,
    "inference_latency_async": 47.777581214904785,
    "inference_throughput_sync": 0.008122611071928475,
    "inference_throughput_async": 0.020930318667702626,
    "inference_CO2_emissions": 1.9401499854764428e-05,
    "inference_energy_consumption": 6.645693546901914e-05
}