{
    "base_current_gpu_type": "NVIDIA A100-PCIE-40GB",
    "base_current_gpu_total_memory": 40339.3125,
    "base_perplexity": 10.554253578186035,
    "base_token_generation_latency_sync": 37.92372055053711,
    "base_token_generation_latency_async": 38.342260010540485,
    "base_token_generation_throughput_sync": 0.02636872082915496,
    "base_token_generation_throughput_async": 0.02608088307066654,
    "base_token_generation_CO2_emissions": null,
    "base_token_generation_energy_consumption": null,
    "base_inference_latency_sync": 120.5372932434082,
    "base_inference_latency_async": 39.461350440979004,
    "base_inference_throughput_sync": 0.00829618762037936,
    "base_inference_throughput_async": 0.025341251346571777,
    "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": 12.661364555358887,
    "smashed_token_generation_latency_sync": 63.71966285705567,
    "smashed_token_generation_latency_async": 62.7002339810133,
    "smashed_token_generation_throughput_sync": 0.015693742797154023,
    "smashed_token_generation_throughput_async": 0.01594890379998928,
    "smashed_token_generation_CO2_emissions": null,
    "smashed_token_generation_energy_consumption": null,
    "smashed_inference_latency_sync": 197.18912048339843,
    "smashed_inference_latency_async": 111.08362674713135,
    "smashed_inference_throughput_sync": 0.005071273696786893,
    "smashed_inference_throughput_async": 0.009002226784298113,
    "smashed_inference_CO2_emissions": null,
    "smashed_inference_energy_consumption": null
}