{
    "base_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "base_current_gpu_total_memory": 40339.3125,
    "base_token_generation_latency_sync": 54.72446517944336,
    "base_token_generation_latency_async": 54.88887540996075,
    "base_token_generation_throughput_sync": 0.018273362685609928,
    "base_token_generation_throughput_async": 0.018218627955684602,
    "base_token_generation_CO2_emissions": null,
    "base_token_generation_energy_consumption": null,
    "base_inference_latency_sync": 54.1922306060791,
    "base_inference_latency_async": 53.47590446472168,
    "base_inference_throughput_sync": 0.018452829655028508,
    "base_inference_throughput_async": 0.01870001096773791,
    "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": 165.96662292480468,
    "smashed_token_generation_latency_async": 166.64433628320694,
    "smashed_token_generation_throughput_sync": 0.006025307874421683,
    "smashed_token_generation_throughput_async": 0.006000804001526525,
    "smashed_token_generation_CO2_emissions": null,
    "smashed_token_generation_energy_consumption": null,
    "smashed_inference_latency_sync": 172.72268676757812,
    "smashed_inference_latency_async": 140.42022228240967,
    "smashed_inference_throughput_sync": 0.00578962740051419,
    "smashed_inference_throughput_async": 0.007121481391681782,
    "smashed_inference_CO2_emissions": null,
    "smashed_inference_energy_consumption": null
}