results.json

{
    "current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "current_gpu_total_memory": 40339.3125,
    "perplexity": 152463.125,
    "token_generation_latency_sync": 175.5655090332031,
    "token_generation_latency_async": 175.71649979799986,
    "token_generation_throughput_sync": 0.005695879592220354,
    "token_generation_throughput_async": 0.005690985201444257,
    "token_generation_CO2_emissions": null,
    "token_generation_energy_consumption": null,
    "inference_latency_sync": 272.8367095947266,
    "inference_latency_async": 234.84888076782227,
    "inference_throughput_sync": 0.0036651959389387393,
    "inference_throughput_async": 0.004258057337682721,
    "inference_CO2_emissions": null,
    "inference_energy_consumption": null
}