{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 1330.3765869140625, "token_generation_latency_sync": 44.9002799987793, "token_generation_latency_async": 45.161220617592335, "token_generation_throughput_sync": 0.022271576035320647, "token_generation_throughput_async": 0.022142891319692427, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 44.401254653930664, "inference_latency_async": 44.11191940307617, "inference_throughput_sync": 0.022521886099708986, "inference_throughput_async": 0.022669609791005022, "inference_CO2_emissions": null, "inference_energy_consumption": null }