| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 82.04161834716797, | |
| "base_token_generation_latency_async": 82.49805271625519, | |
| "base_token_generation_throughput_sync": 0.012188935568876664, | |
| "base_token_generation_throughput_async": 0.012121498230260201, | |
| "base_token_generation_CO2_emissions": 2.1165249602943306e-05, | |
| "base_token_generation_energy_consumption": 0.006756192727641794, | |
| "base_inference_latency_sync": 81.24088439941406, | |
| "base_inference_latency_async": 78.74789237976074, | |
| "base_inference_throughput_sync": 0.012309073287331327, | |
| "base_inference_throughput_async": 0.012698752560608382, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_token_generation_latency_sync": 180.8959991455078, | |
| "smashed_token_generation_latency_async": 181.48778304457664, | |
| "smashed_token_generation_throughput_sync": 0.005528038235912709, | |
| "smashed_token_generation_throughput_async": 0.005510012758017889, | |
| "smashed_token_generation_CO2_emissions": 5.7140606324742935e-05, | |
| "smashed_token_generation_energy_consumption": 0.019805487794740197, | |
| "smashed_inference_latency_sync": 188.26690673828125, | |
| "smashed_inference_latency_async": 162.53809928894043, | |
| "smashed_inference_throughput_sync": 0.005311607957686092, | |
| "smashed_inference_throughput_async": 0.006152403678735789 | |
| } |