| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.405904769897461, | |
| "base_token_generation_latency_sync": 37.993507385253906, | |
| "base_token_generation_latency_async": 39.05783258378506, | |
| "base_token_generation_throughput_sync": 0.026320286512640352, | |
| "base_token_generation_throughput_async": 0.025603059203422162, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 118.9781494140625, | |
| "base_inference_latency_async": 38.94762992858887, | |
| "base_inference_throughput_sync": 0.008404904639421179, | |
| "base_inference_throughput_async": 0.025675503280520966, | |
| "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": 37720.5078125, | |
| "smashed_token_generation_latency_sync": 170.02661895751953, | |
| "smashed_token_generation_latency_async": 170.09772770106792, | |
| "smashed_token_generation_throughput_sync": 0.005881432014182709, | |
| "smashed_token_generation_throughput_async": 0.0058789733026734705, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 265.8291702270508, | |
| "smashed_inference_latency_async": 212.9425287246704, | |
| "smashed_inference_throughput_sync": 0.0037618143981184497, | |
| "smashed_inference_throughput_async": 0.004696102774720855, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |