{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 32063.57421875, "token_generation_latency_sync": 5.327481031417847, "token_generation_latency_async": 5.42500726878643, "token_generation_throughput_sync": 0.18770597100255873, "token_generation_throughput_async": 0.18433155025499148, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 4.612608003616333, "inference_latency_async": 4.320836067199707, "inference_throughput_sync": 0.2167970916271206, "inference_throughput_async": 0.23143669059587593, "inference_CO2_emissions": null, "inference_energy_consumption": null }