| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 9.401168823242188, | |
| "base_token_generation_latency_sync": 43.53713188171387, | |
| "base_token_generation_latency_async": 40.53729362785816, | |
| "base_token_generation_throughput_sync": 0.022968899345894037, | |
| "base_token_generation_throughput_async": 0.024668642390886623, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 121.48449325561523, | |
| "base_inference_latency_async": 44.87926959991455, | |
| "base_inference_throughput_sync": 0.008231503241289424, | |
| "base_inference_throughput_async": 0.02228200255741025, | |
| "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": 117302.8125, | |
| "smashed_token_generation_latency_sync": 172.76212005615236, | |
| "smashed_token_generation_latency_async": 174.94924943894148, | |
| "smashed_token_generation_throughput_sync": 0.00578830590684447, | |
| "smashed_token_generation_throughput_async": 0.0057159433561846, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 265.6924682617188, | |
| "smashed_inference_latency_async": 212.2485637664795, | |
| "smashed_inference_throughput_sync": 0.0037637498967978125, | |
| "smashed_inference_throughput_async": 0.004711457087173612, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |