{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 13287.5927734375, "token_generation_latency_sync": 299.3733215332031, "token_generation_latency_async": 299.73516799509525, "token_generation_throughput_sync": 0.0033403110032604933, "token_generation_throughput_async": 0.003336278511090042, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 463.15724792480466, "inference_latency_async": 415.75257778167725, "inference_throughput_sync": 0.002159093924321689, "inference_throughput_async": 0.0024052767281340266, "inference_CO2_emissions": null, "inference_energy_consumption": null }