{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 133522.1875, "token_generation_latency_sync": 92.61939010620117, "token_generation_latency_async": 92.80381612479687, "token_generation_throughput_sync": 0.010796875242358638, "token_generation_throughput_async": 0.010775418961815768, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 92.3891716003418, "inference_latency_async": 93.42858791351318, "inference_throughput_sync": 0.010823779266317184, "inference_throughput_async": 0.010703362025825539, "inference_CO2_emissions": null, "inference_energy_consumption": null }