{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 9.208965301513672, "token_generation_latency_sync": 11.42576847076416, "token_generation_latency_async": 11.491917446255684, "token_generation_throughput_sync": 0.08752146541029285, "token_generation_throughput_async": 0.08701768044164133, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 49.50845527648926, "inference_latency_async": 21.486544609069824, "inference_throughput_sync": 0.020198570010219716, "inference_throughput_async": 0.04654075460685678, "inference_CO2_emissions": null, "inference_energy_consumption": null }