{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 207388.0, "token_generation_latency_sync": 174.94263610839843, "token_generation_latency_async": 175.12229178100824, "token_generation_throughput_sync": 0.005716159435144085, "token_generation_throughput_async": 0.005710295301814047, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 271.9763442993164, "inference_latency_async": 233.81962776184082, "inference_throughput_sync": 0.0036767903568094007, "inference_throughput_async": 0.004276800923738359, "inference_CO2_emissions": null, "inference_energy_consumption": null }