{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 34.19964599609375, "token_generation_latency_sync": 16.38497323989868, "token_generation_latency_async": 16.560293547809124, "token_generation_throughput_sync": 0.06103153086420199, "token_generation_throughput_async": 0.06038540301915704, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 184.8563720703125, "inference_latency_async": 49.31023120880127, "inference_throughput_sync": 0.005409605245415274, "inference_throughput_async": 0.02027976700749098, "inference_CO2_emissions": null, "inference_energy_consumption": null }