{
    "base_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "base_current_gpu_total_memory": 40339.3125,
    "base_perplexity": 82.86221313476562,
    "base_token_generation_latency_sync": 23.3876594543457,
    "base_token_generation_latency_async": 23.057953640818596,
    "base_token_generation_throughput_sync": 0.042757591966484196,
    "base_token_generation_throughput_async": 0.043368983023269636,
    "base_token_generation_CO2_emissions": null,
    "base_token_generation_energy_consumption": null,
    "base_inference_latency_sync": 41.338264846801756,
    "base_inference_latency_async": 23.22986125946045,
    "base_inference_throughput_sync": 0.024190662179604465,
    "base_inference_throughput_async": 0.043048040142415665,
    "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": 82.88560485839844,
    "smashed_token_generation_latency_sync": 38.0381778717041,
    "smashed_token_generation_latency_async": 38.69975656270981,
    "smashed_token_generation_throughput_sync": 0.026289377040425523,
    "smashed_token_generation_throughput_async": 0.02583995582451743,
    "smashed_token_generation_CO2_emissions": null,
    "smashed_token_generation_energy_consumption": null,
    "smashed_inference_latency_sync": 66.39616012573242,
    "smashed_inference_latency_async": 47.6445198059082,
    "smashed_inference_throughput_sync": 0.01506111193940026,
    "smashed_inference_throughput_async": 0.02098877277121794,
    "smashed_inference_CO2_emissions": null,
    "smashed_inference_energy_consumption": null
}