| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.405904769897461, | |
| "base_token_generation_latency_sync": 38.60344696044922, | |
| "base_token_generation_latency_async": 38.134160824120045, | |
| "base_token_generation_throughput_sync": 0.025904422499486643, | |
| "base_token_generation_throughput_async": 0.026223207181931614, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 120.48721923828126, | |
| "base_inference_latency_async": 39.444828033447266, | |
| "base_inference_throughput_sync": 0.008299635482684288, | |
| "base_inference_throughput_async": 0.025351866134440983, | |
| "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": 9.43250846862793, | |
| "smashed_token_generation_latency_sync": 62.9321174621582, | |
| "smashed_token_generation_latency_async": 62.363552674651146, | |
| "smashed_token_generation_throughput_sync": 0.015890137505722914, | |
| "smashed_token_generation_throughput_async": 0.016035006940944674, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 198.10078887939454, | |
| "smashed_inference_latency_async": 111.86470985412598, | |
| "smashed_inference_throughput_sync": 0.005047935475960212, | |
| "smashed_inference_throughput_async": 0.008939369719941363, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |