{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 9.303628921508789, "token_generation_latency_sync": 291.45216064453126, "token_generation_latency_async": 291.4591655135155, "token_generation_throughput_sync": 0.003431094824579623, "token_generation_throughput_async": 0.003431012362359997, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 279.33388671875, "inference_latency_async": 195.6489086151123, "inference_throughput_sync": 0.003579945175097426, "inference_throughput_async": 0.005111196413404158, "inference_CO2_emissions": null, "inference_energy_consumption": null }