{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": NaN, "token_generation_latency_sync": 45.5171501159668, "token_generation_latency_async": 45.524014718830585, "token_generation_throughput_sync": 0.021969741019642913, "token_generation_throughput_async": 0.02196642818469082, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 214.7057662963867, "inference_latency_async": 108.77645015716553, "inference_throughput_sync": 0.004657536764148048, "inference_throughput_async": 0.009193166338441373, "inference_CO2_emissions": null, "inference_energy_consumption": null }