{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 171741.578125, "token_generation_latency_sync": 107.70650711059571, "token_generation_latency_async": 105.71706611663103, "token_generation_throughput_sync": 0.009284490109526766, "token_generation_throughput_async": 0.00945921067178276, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 110.17267303466797, "inference_latency_async": 108.40392112731934, "inference_throughput_sync": 0.009076660958251695, "inference_throughput_async": 0.00922475856593333, "inference_CO2_emissions": null, "inference_energy_consumption": null }