| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 55.28608436584473, | |
| "base_token_generation_latency_async": 54.56864908337593, | |
| "base_token_generation_throughput_sync": 0.018087734218663377, | |
| "base_token_generation_throughput_async": 0.01832554070510507, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 53.28803901672363, | |
| "base_inference_latency_async": 53.00874710083008, | |
| "base_inference_throughput_sync": 0.018765937318244445, | |
| "base_inference_throughput_async": 0.018864811086703476, | |
| "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": 165.7446533203125, | |
| "smashed_token_generation_latency_async": 166.8392814695835, | |
| "smashed_token_generation_throughput_sync": 0.0060333771253992365, | |
| "smashed_token_generation_throughput_async": 0.0059937922963442525, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 173.7375717163086, | |
| "smashed_inference_latency_async": 141.0609483718872, | |
| "smashed_inference_throughput_sync": 0.005755807394573657, | |
| "smashed_inference_throughput_async": 0.007089134246876334, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |