| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 82.86221313476562, | |
| "base_token_generation_latency_sync": 22.66096477508545, | |
| "base_token_generation_latency_async": 20.442129857838154, | |
| "base_token_generation_throughput_sync": 0.04412874782363405, | |
| "base_token_generation_throughput_async": 0.048918581720904616, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 41.06926002502441, | |
| "base_inference_latency_async": 18.918275833129883, | |
| "base_inference_throughput_sync": 0.024349111705218886, | |
| "base_inference_throughput_async": 0.05285893962116725, | |
| "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": Infinity, | |
| "smashed_token_generation_latency_sync": 85.1107421875, | |
| "smashed_token_generation_latency_async": 82.53523223102093, | |
| "smashed_token_generation_throughput_sync": 0.011749398187563537, | |
| "smashed_token_generation_throughput_async": 0.012116037878235342, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 88.11602020263672, | |
| "smashed_inference_latency_async": 77.12936401367188, | |
| "smashed_inference_throughput_sync": 0.011348674142344852, | |
| "smashed_inference_throughput_async": 0.012965230723576833, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |