| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 51.796264266967775, | |
| "base_token_generation_latency_async": 51.505282148718834, | |
| "base_token_generation_throughput_sync": 0.019306411652504712, | |
| "base_token_generation_throughput_async": 0.0194154843596925, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 50.28413467407226, | |
| "base_inference_latency_async": 48.9729642868042, | |
| "base_inference_throughput_sync": 0.019886988340989082, | |
| "base_inference_throughput_async": 0.020419429670289545, | |
| "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": 178.5696228027344, | |
| "smashed_token_generation_latency_async": 178.96188646554947, | |
| "smashed_token_generation_throughput_sync": 0.005600056629478905, | |
| "smashed_token_generation_throughput_async": 0.0055877819559781074, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 185.12169036865234, | |
| "smashed_inference_latency_async": 132.38873481750488, | |
| "smashed_inference_throughput_sync": 0.005401852143898398, | |
| "smashed_inference_throughput_async": 0.007553512777189684, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |