| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 52.6988166809082, | |
| "base_token_generation_latency_async": 52.70070154219866, | |
| "base_token_generation_throughput_sync": 0.01897575814756921, | |
| "base_token_generation_throughput_async": 0.018975079472125757, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 52.19194869995117, | |
| "base_inference_latency_async": 51.29694938659668, | |
| "base_inference_throughput_sync": 0.01916004335743332, | |
| "base_inference_throughput_async": 0.019494336641026237, | |
| "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": 163.9918991088867, | |
| "smashed_token_generation_latency_async": 164.94027692824602, | |
| "smashed_token_generation_throughput_sync": 0.006097862183643741, | |
| "smashed_token_generation_throughput_async": 0.006062800539828305, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 172.04705352783202, | |
| "smashed_inference_latency_async": 140.13025760650635, | |
| "smashed_inference_throughput_sync": 0.005812363417419585, | |
| "smashed_inference_throughput_async": 0.007136217524184222, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |