{
    "base_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "base_current_gpu_total_memory": 40339.3125,
    "base_token_generation_latency_sync": 53.123374938964844,
    "base_token_generation_latency_async": 53.1091520562768,
    "base_token_generation_throughput_sync": 0.01882410522955163,
    "base_token_generation_throughput_async": 0.018829146414168992,
    "base_token_generation_CO2_emissions": null,
    "base_token_generation_energy_consumption": null,
    "base_inference_latency_sync": 52.437298965454104,
    "base_inference_latency_async": 51.26469135284424,
    "base_inference_throughput_sync": 0.019070394923636397,
    "base_inference_throughput_async": 0.019506603348437377,
    "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": 167.08106079101563,
    "smashed_token_generation_latency_async": 167.67955236136913,
    "smashed_token_generation_throughput_sync": 0.005985118811585691,
    "smashed_token_generation_throughput_async": 0.005963756378862955,
    "smashed_token_generation_CO2_emissions": null,
    "smashed_token_generation_energy_consumption": null,
    "smashed_inference_latency_sync": 174.62354125976563,
    "smashed_inference_latency_async": 141.94059371948242,
    "smashed_inference_throughput_sync": 0.005726604745189681,
    "smashed_inference_throughput_async": 0.007045200909729198,
    "smashed_inference_CO2_emissions": null,
    "smashed_inference_energy_consumption": null
}