base_results.json

{
    "current_gpu_type": "NVIDIA L40S",
    "current_gpu_total_memory": 45372.6875,
    "memory_inference_first": 25590.0,
    "memory_inference": 25590.0,
    "token_generation_latency_sync": 30.846586608886717,
    "token_generation_latency_async": 30.805427208542824,
    "token_generation_throughput_sync": 0.03241849779618423,
    "token_generation_throughput_async": 0.03246181243422862,
    "token_generation_CO2_emissions": 1.7413158175960427e-06,
    "token_generation_energy_consumption": 0.0012653706804628014,
    "inference_latency_sync": 25.427967834472657,
    "inference_latency_async": 24.5699405670166,
    "inference_throughput_sync": 0.039326776190282166,
    "inference_throughput_async": 0.04070013915061842,
    "inference_CO2_emissions": 2.974451769021189e-06,
    "inference_energy_consumption": 1.0356497508141638e-05
}