| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 105.3484390258789, | |
| "base_token_generation_latency_async": 104.91261184215546, | |
| "base_token_generation_throughput_sync": 0.009492309608444691, | |
| "base_token_generation_throughput_async": 0.009531742489687833, | |
| "base_token_generation_CO2_emissions": 2.412649618989479e-05, | |
| "base_token_generation_energy_consumption": 0.008419673305905371, | |
| "base_inference_latency_sync": 86.63613433837891, | |
| "base_inference_latency_async": 88.81196975708008, | |
| "base_inference_throughput_sync": 0.01154252792598354, | |
| "base_inference_throughput_async": 0.01125974350906996, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_token_generation_latency_sync": 57.02575531005859, | |
| "smashed_token_generation_latency_async": 55.63120283186436, | |
| "smashed_token_generation_throughput_sync": 0.017535936079457998, | |
| "smashed_token_generation_throughput_async": 0.017975523610775167, | |
| "smashed_token_generation_CO2_emissions": 1.8021672618399667e-05, | |
| "smashed_token_generation_energy_consumption": 0.004920462357914908, | |
| "smashed_inference_latency_sync": 59.50504913330078, | |
| "smashed_inference_latency_async": 57.179927825927734, | |
| "smashed_inference_throughput_sync": 0.016805296601971387, | |
| "smashed_inference_throughput_async": 0.01748865446357837 | |
| } |