{ "current_gpu_type": "Tesla T4", "current_gpu_total_memory": 15095.0625, "perplexity": 3.8198323249816895, "memory_inference_first": 206.0, "memory_inference": 206.0, "token_generation_latency_sync": 57.987845993041994, "token_generation_latency_async": 58.05423278361559, "token_generation_throughput_sync": 0.017244993030435907, "token_generation_throughput_async": 0.01722527285352785, "token_generation_CO2_emissions": 9.527744243728705e-06, "token_generation_energy_consumption": 0.003671008174955222, "inference_latency_sync": 58.546810150146484, "inference_latency_async": 54.94275093078613, "inference_throughput_sync": 0.017080349850580167, "inference_throughput_async": 0.018200763213690286, "inference_CO2_emissions": 9.524661982723253e-06, "inference_energy_consumption": 3.420485454644485e-05 }