| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.629734992980957, | |
| "base_token_generation_latency_sync": 38.651375961303714, | |
| "base_token_generation_latency_async": 38.82457036525011, | |
| "base_token_generation_throughput_sync": 0.025872300147895433, | |
| "base_token_generation_throughput_async": 0.02575688515268282, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.93804702758788, | |
| "base_inference_latency_async": 39.46812152862549, | |
| "base_inference_throughput_sync": 0.00833763784539515, | |
| "base_inference_throughput_async": 0.025336903842122782, | |
| "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": 9.242453575134277, | |
| "smashed_token_generation_latency_sync": 165.535498046875, | |
| "smashed_token_generation_latency_async": 165.71880243718624, | |
| "smashed_token_generation_throughput_sync": 0.00604100034010124, | |
| "smashed_token_generation_throughput_async": 0.006034318286719687, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 263.5079681396484, | |
| "smashed_inference_latency_async": 194.88348960876465, | |
| "smashed_inference_throughput_sync": 0.0037949516557694413, | |
| "smashed_inference_throughput_async": 0.005131271007141419, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |