| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.891793251037598, | |
| "base_token_generation_latency_sync": 35.77095603942871, | |
| "base_token_generation_latency_async": 35.82000117748976, | |
| "base_token_generation_throughput_sync": 0.027955640852812127, | |
| "base_token_generation_throughput_async": 0.027917363683070635, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 118.29401626586915, | |
| "base_inference_latency_async": 38.65203857421875, | |
| "base_inference_throughput_sync": 0.008453512963432332, | |
| "base_inference_throughput_async": 0.025871856618372744, | |
| "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_perplexity": 9.283415794372559, | |
| "smashed_token_generation_latency_sync": 41.99628257751465, | |
| "smashed_token_generation_latency_async": 42.51692555844784, | |
| "smashed_token_generation_throughput_sync": 0.02381163137842617, | |
| "smashed_token_generation_throughput_async": 0.023520044943637897, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 162.3283706665039, | |
| "smashed_inference_latency_async": 70.35641670227051, | |
| "smashed_inference_throughput_sync": 0.006160352598218666, | |
| "smashed_inference_throughput_async": 0.014213344665231203, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |