| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.711685180664062, | |
| "base_token_generation_latency_sync": 38.78611068725586, | |
| "base_token_generation_latency_async": 38.81423454731703, | |
| "base_token_generation_throughput_sync": 0.025782425262055854, | |
| "base_token_generation_throughput_async": 0.02576374393731599, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 118.54264373779297, | |
| "base_inference_latency_async": 39.09869194030762, | |
| "base_inference_throughput_sync": 0.008435782841252652, | |
| "base_inference_throughput_async": 0.02557630320540417, | |
| "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": 210114.78125, | |
| "smashed_token_generation_latency_sync": 168.86386260986328, | |
| "smashed_token_generation_latency_async": 169.0822370350361, | |
| "smashed_token_generation_throughput_sync": 0.0059219301545314195, | |
| "smashed_token_generation_throughput_async": 0.0059142818165623555, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 265.30672760009764, | |
| "smashed_inference_latency_async": 212.2807502746582, | |
| "smashed_inference_throughput_sync": 0.003769222171807572, | |
| "smashed_inference_throughput_async": 0.004710742724934578, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |