{
    "base_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "base_current_gpu_total_memory": 40339.3125,
    "base_perplexity": 6.637787342071533,
    "base_token_generation_latency_sync": 40.0831859588623,
    "base_token_generation_latency_async": 41.7502811178565,
    "base_token_generation_throughput_sync": 0.024948116674814922,
    "base_token_generation_throughput_async": 0.023951934531341448,
    "base_token_generation_CO2_emissions": null,
    "base_token_generation_energy_consumption": null,
    "base_inference_latency_sync": 120.26081161499023,
    "base_inference_latency_async": 39.661383628845215,
    "base_inference_throughput_sync": 0.008315260695241744,
    "base_inference_throughput_async": 0.025213442106762834,
    "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": 7.3032684326171875,
    "smashed_token_generation_latency_sync": 166.38088989257812,
    "smashed_token_generation_latency_async": 166.69232975691557,
    "smashed_token_generation_throughput_sync": 0.006010305634533138,
    "smashed_token_generation_throughput_async": 0.005999076270985486,
    "smashed_token_generation_CO2_emissions": null,
    "smashed_token_generation_energy_consumption": null,
    "smashed_inference_latency_sync": 263.4641403198242,
    "smashed_inference_latency_async": 195.7735300064087,
    "smashed_inference_throughput_sync": 0.00379558295404483,
    "smashed_inference_throughput_async": 0.005107942835618609,
    "smashed_inference_CO2_emissions": null,
    "smashed_inference_energy_consumption": null
}