{
    "base_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "base_current_gpu_total_memory": 40339.3125,
    "base_perplexity": 13.74486255645752,
    "base_token_generation_latency_sync": 37.35860481262207,
    "base_token_generation_latency_async": 37.739571928977966,
    "base_token_generation_throughput_sync": 0.026767594909276098,
    "base_token_generation_throughput_async": 0.026497385870775065,
    "base_token_generation_CO2_emissions": null,
    "base_token_generation_energy_consumption": null,
    "base_inference_latency_sync": 122.0685821533203,
    "base_inference_latency_async": 43.52927207946777,
    "base_inference_throughput_sync": 0.008192116123246046,
    "base_inference_throughput_async": 0.022973046693139806,
    "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": 14.672628402709961,
    "smashed_token_generation_latency_sync": 43.5498722076416,
    "smashed_token_generation_latency_async": 43.305750377476215,
    "smashed_token_generation_throughput_sync": 0.02296217989417044,
    "smashed_token_generation_throughput_async": 0.02309162158104783,
    "smashed_token_generation_CO2_emissions": null,
    "smashed_token_generation_energy_consumption": null,
    "smashed_inference_latency_sync": 164.25011291503907,
    "smashed_inference_latency_async": 69.53554153442383,
    "smashed_inference_throughput_sync": 0.006088275875446525,
    "smashed_inference_throughput_async": 0.014381134854683576,
    "smashed_inference_CO2_emissions": null,
    "smashed_inference_energy_consumption": null
}