{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 12.936248779296875, "token_generation_latency_sync": 17.07958984375, "token_generation_latency_async": 17.199922166764736, "token_generation_throughput_sync": 0.05854941536350382, "token_generation_throughput_async": 0.05813979797724268, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 189.4657012939453, "inference_latency_async": 47.06454277038574, "inference_throughput_sync": 0.005278000150795402, "inference_throughput_async": 0.021247417719082284, "inference_CO2_emissions": null, "inference_energy_consumption": null }