{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 131872.640625, "token_generation_latency_sync": 282.989924621582, "token_generation_latency_async": 283.4888217970729, "token_generation_throughput_sync": 0.0035336947113478105, "token_generation_throughput_async": 0.0035274759465324544, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 273.77950897216795, "inference_latency_async": 232.7092409133911, "inference_throughput_sync": 0.0036525743060692633, "inference_throughput_async": 0.0042972079496068506, "inference_CO2_emissions": null, "inference_energy_consumption": null }