results.json

{
    "current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "current_gpu_total_memory": 40339.3125,
    "perplexity": 7.029494285583496,
    "token_generation_latency_sync": 49.039611434936525,
    "token_generation_latency_async": 48.72753079980612,
    "token_generation_throughput_sync": 0.02039167870093656,
    "token_generation_throughput_async": 0.020522279368278166,
    "token_generation_CO2_emissions": null,
    "token_generation_energy_consumption": null,
    "inference_latency_sync": 189.3416946411133,
    "inference_latency_async": 102.76260375976562,
    "inference_throughput_sync": 0.0052814569020069495,
    "inference_throughput_async": 0.009731166430326742,
    "inference_CO2_emissions": null,
    "inference_energy_consumption": null
}