{
    "base_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "base_current_gpu_total_memory": 40339.3125,
    "base_perplexity": 11.575923919677734,
    "base_token_generation_latency_sync": 36.300475692749025,
    "base_token_generation_latency_async": 36.70228775590658,
    "base_token_generation_throughput_sync": 0.02754784836606835,
    "base_token_generation_throughput_async": 0.02724625796219114,
    "base_token_generation_CO2_emissions": null,
    "base_token_generation_energy_consumption": null,
    "base_inference_latency_sync": 118.6505744934082,
    "base_inference_latency_async": 38.658761978149414,
    "base_inference_throughput_sync": 0.008428109212868214,
    "base_inference_throughput_async": 0.025867357070700217,
    "base_inference_CO2_emissions": null,
    "base_inference_energy_consumption": null,
    "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "smashed_current_gpu_total_memory": 40339.3125,
    "smashed_perplexity": 11.946749687194824,
    "smashed_token_generation_latency_sync": 47.3871337890625,
    "smashed_token_generation_latency_async": 47.25001659244299,
    "smashed_token_generation_throughput_sync": 0.02110277453055858,
    "smashed_token_generation_throughput_async": 0.021164013732006534,
    "smashed_token_generation_CO2_emissions": null,
    "smashed_token_generation_energy_consumption": null,
    "smashed_inference_latency_sync": 161.72431335449218,
    "smashed_inference_latency_async": 70.20044326782227,
    "smashed_inference_throughput_sync": 0.006183362162793954,
    "smashed_inference_throughput_async": 0.014244924297484734,
    "smashed_inference_CO2_emissions": null,
    "smashed_inference_energy_consumption": null
}