{
    "base_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "base_current_gpu_total_memory": 40339.3125,
    "base_token_generation_latency_sync": 55.15896377563477,
    "base_token_generation_latency_async": 55.11929299682379,
    "base_token_generation_throughput_sync": 0.01812941961831646,
    "base_token_generation_throughput_async": 0.018142467829869013,
    "base_token_generation_CO2_emissions": null,
    "base_token_generation_energy_consumption": null,
    "base_inference_latency_sync": 54.881177520751955,
    "base_inference_latency_async": 55.046939849853516,
    "base_inference_throughput_sync": 0.018221183385175633,
    "base_inference_throughput_async": 0.01816631410806138,
    "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_token_generation_latency_sync": 173.44444580078124,
    "smashed_token_generation_latency_async": 173.68598524481058,
    "smashed_token_generation_throughput_sync": 0.005765534868430452,
    "smashed_token_generation_throughput_async": 0.005757516926829179,
    "smashed_token_generation_CO2_emissions": null,
    "smashed_token_generation_energy_consumption": null,
    "smashed_inference_latency_sync": 179.36209869384766,
    "smashed_inference_latency_async": 156.9817066192627,
    "smashed_inference_throughput_sync": 0.005575313888955411,
    "smashed_inference_throughput_async": 0.006370168993164032,
    "smashed_inference_CO2_emissions": null,
    "smashed_inference_energy_consumption": null
}