{ "current_gpu_type": "Tesla T4", "current_gpu_total_memory": 15095.0625, "perplexity": 3.4586403369903564, "memory_inference_first": 808.0, "memory_inference": 808.0, "token_generation_latency_sync": 36.94038696289063, "token_generation_latency_async": 37.75534201413393, "token_generation_throughput_sync": 0.027070642248674184, "token_generation_throughput_async": 0.026486318138123188, "token_generation_CO2_emissions": 2.9548104595618817e-05, "token_generation_energy_consumption": 0.0024448525932904515, "inference_latency_sync": 259.3957290649414, "inference_latency_async": 82.87992477416992, "inference_throughput_sync": 0.0038551135888195117, "inference_throughput_async": 0.012065648016993093, "inference_CO2_emissions": 3.8677334298918255e-05, "inference_energy_consumption": 0.00013983032998305092 }