{ "current_gpu_type": "Tesla T4", "current_gpu_total_memory": 15095.0625, "perplexity": 343688032.0, "memory_inference_first": 344.0, "memory_inference": 344.0, "token_generation_latency_sync": 94.19695053100585, "token_generation_latency_async": 93.61793361604214, "token_generation_throughput_sync": 0.010616054918580831, "token_generation_throughput_async": 0.01068171408377083, "token_generation_CO2_emissions": 2.0701273057098414e-05, "token_generation_energy_consumption": 0.004600785530736977, "inference_latency_sync": 146.6492561340332, "inference_latency_async": 93.28432083129883, "inference_throughput_sync": 0.006818991288206936, "inference_throughput_async": 0.010719915105652773, "inference_CO2_emissions": 2.082240367862723e-05, "inference_energy_consumption": 7.15078451615474e-05 }