| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 13.74486255645752, | |
| "base_token_generation_latency_sync": 32.37070541381836, | |
| "base_token_generation_latency_async": 32.25280176848173, | |
| "base_token_generation_throughput_sync": 0.030892128769400295, | |
| "base_token_generation_throughput_async": 0.031005058325730504, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 123.61564178466797, | |
| "base_inference_latency_async": 43.86191368103027, | |
| "base_inference_throughput_sync": 0.008089591135577714, | |
| "base_inference_throughput_async": 0.0227988228528316, | |
| "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": 170382.984375, | |
| "smashed_token_generation_latency_sync": 175.3365936279297, | |
| "smashed_token_generation_latency_async": 175.29774066060781, | |
| "smashed_token_generation_throughput_sync": 0.005703316001005669, | |
| "smashed_token_generation_throughput_async": 0.0057045800831859545, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 270.7505142211914, | |
| "smashed_inference_latency_async": 230.96849918365479, | |
| "smashed_inference_throughput_sync": 0.003693437121907157, | |
| "smashed_inference_throughput_async": 0.0043295947435881684, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |