| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 11.575923919677734, | |
| "base_token_generation_latency_sync": 38.96055145263672, | |
| "base_token_generation_latency_async": 38.49546890705824, | |
| "base_token_generation_throughput_sync": 0.025666987830387174, | |
| "base_token_generation_throughput_async": 0.025977083235805123, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.22933731079101, | |
| "base_inference_latency_async": 39.063167572021484, | |
| "base_inference_throughput_sync": 0.008387197501512018, | |
| "base_inference_throughput_async": 0.025599562507476678, | |
| "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": 174225.5, | |
| "smashed_token_generation_latency_sync": 179.29608459472655, | |
| "smashed_token_generation_latency_async": 174.7506869956851, | |
| "smashed_token_generation_throughput_sync": 0.005577366634973421, | |
| "smashed_token_generation_throughput_async": 0.005722438161428755, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 267.67871856689453, | |
| "smashed_inference_latency_async": 230.59861660003662, | |
| "smashed_inference_throughput_sync": 0.0037358218290711592, | |
| "smashed_inference_throughput_async": 0.004336539458666645, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |