| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 56.18572578430176, | |
| "base_token_generation_latency_async": 55.955370515584946, | |
| "base_token_generation_throughput_sync": 0.017798114842175788, | |
| "base_token_generation_throughput_async": 0.017871385548621745, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 54.89110984802246, | |
| "base_inference_latency_async": 53.778696060180664, | |
| "base_inference_throughput_sync": 0.018217886334758206, | |
| "base_inference_throughput_async": 0.018594723808122035, | |
| "base_inference_CO2_emissions": null, | |
| "base_inference_energy_consumption": null, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_token_generation_latency_sync": 163.71141204833984, | |
| "smashed_token_generation_latency_async": 164.27245195955038, | |
| "smashed_token_generation_throughput_sync": 0.006108309662033367, | |
| "smashed_token_generation_throughput_async": 0.0060874479443834865, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 171.10599670410156, | |
| "smashed_inference_latency_async": 128.42419147491455, | |
| "smashed_inference_throughput_sync": 0.005844330527639708, | |
| "smashed_inference_throughput_async": 0.007786694924961491, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |