{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 23332.130859375, "token_generation_latency_sync": 274.0644332885742, "token_generation_latency_async": 274.208939447999, "token_generation_throughput_sync": 0.0036487769974408064, "token_generation_throughput_async": 0.0036468541179330884, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 263.15919342041013, "inference_latency_async": 210.33341884613037, "inference_throughput_sync": 0.003799981247101823, "inference_throughput_async": 0.00475435622872441, "inference_CO2_emissions": null, "inference_energy_consumption": null }