| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 83.92937774658203, | |
| "base_token_generation_latency_async": 83.90798717737198, | |
| "base_token_generation_throughput_sync": 0.011914779149434649, | |
| "base_token_generation_throughput_async": 0.011917816570740916, | |
| "base_token_generation_CO2_emissions": 2.0898577111994714e-05, | |
| "base_token_generation_energy_consumption": 0.006837025531716293, | |
| "base_inference_latency_sync": 81.4477310180664, | |
| "base_inference_latency_async": 80.29217720031738, | |
| "base_inference_throughput_sync": 0.012277812868454053, | |
| "base_inference_throughput_async": 0.012454513439150423, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_token_generation_latency_sync": 209.10842590332032, | |
| "smashed_token_generation_latency_async": 208.9206464588642, | |
| "smashed_token_generation_throughput_sync": 0.004782208061105784, | |
| "smashed_token_generation_throughput_async": 0.004786506345589432, | |
| "smashed_token_generation_CO2_emissions": 6.253313396231958e-05, | |
| "smashed_token_generation_energy_consumption": 0.0220396598549638, | |
| "smashed_inference_latency_sync": 206.4805877685547, | |
| "smashed_inference_latency_async": 205.32312393188477, | |
| "smashed_inference_throughput_sync": 0.004843070289595, | |
| "smashed_inference_throughput_async": 0.004870372030437968 | |
| } |