| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 31.19168758392334, | |
| "base_token_generation_latency_async": 31.13317396491766, | |
| "base_token_generation_throughput_sync": 0.03205982354463613, | |
| "base_token_generation_throughput_async": 0.03212007876636181, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 30.637567901611327, | |
| "base_inference_latency_async": 29.344534873962402, | |
| "base_inference_throughput_sync": 0.0326396665430942, | |
| "base_inference_throughput_async": 0.034077895740896764, | |
| "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": 71.31928787231445, | |
| "smashed_token_generation_latency_async": 70.39164807647467, | |
| "smashed_token_generation_throughput_sync": 0.014021452398548018, | |
| "smashed_token_generation_throughput_async": 0.01420623081467823, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 68.66606101989746, | |
| "smashed_inference_latency_async": 67.55332946777344, | |
| "smashed_inference_throughput_sync": 0.014563235245287022, | |
| "smashed_inference_throughput_async": 0.014803119370704795, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |