| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 10.639251708984375, | |
| "base_token_generation_latency_sync": 39.957108688354495, | |
| "base_token_generation_latency_async": 39.34535309672356, | |
| "base_token_generation_throughput_sync": 0.025026835845393645, | |
| "base_token_generation_throughput_async": 0.025415962020767173, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.04286651611328, | |
| "base_inference_latency_async": 40.1911735534668, | |
| "base_inference_throughput_sync": 0.008400335352010723, | |
| "base_inference_throughput_async": 0.024881084864807147, | |
| "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": 11.474609375, | |
| "smashed_token_generation_latency_sync": 47.31715698242188, | |
| "smashed_token_generation_latency_async": 46.08920980244875, | |
| "smashed_token_generation_throughput_sync": 0.021133983184397485, | |
| "smashed_token_generation_throughput_async": 0.02169705239656483, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 162.52672119140624, | |
| "smashed_inference_latency_async": 70.94693183898926, | |
| "smashed_inference_throughput_sync": 0.006152834393442965, | |
| "smashed_inference_throughput_async": 0.014095042224933041, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |